Modeling the Effect of Water Activity on Thermal Resistance of Salmonella in Wheat Flour

MODELING THE EFFECT OF WATER ACTIVITY ON THERMAL RESISTANCE OF SALMONELLA IN WHEAT FLOUR

Danielle F. Smith

A THESIS

Submitted to Michigan State University in partial fulfillment of the requirements for the degree of

Biosystems Engineering – Master of Science

2014

ABSTRACT

MODELING THE EFFECT OF WATER ACTIVITY ON THERMAL RESISTANCE OF SALMONELLA IN WHEAT FLOUR

Danielle F. Smith

Salmonella is able to survive in low-moisture environments, is known to be more heat resistant as product water activity decreases, and recently has been implicated in several outbreaks and/or recalls associated with low-moisture foods. Therefore, the specific objectives were to: (1) Evaluate the effect of rapid product desiccation and hydration on the thermal resistance of Salmonella Enteritidis PT 30 in wheat flour, and (2) Test multiple secondary models for the effect of product (wheat flour) water activity on Salmonella thermal resistance. A custom-built fluidized-bed drying and hydrating system was used to rapidly (1-4 min) change the sample’s water activity (0.6 to 0.3, or 0.3 to 0.6) prior to thermal treatment, and the results were compared with those from samples with longer equilibration times (4-7 days). Desiccation or hydration rate did not affect Salmonella thermal resistance; instead, product water activity at the time of thermal treatment controlled Salmonella thermal resistance, regardless of the water activity “history”. An additional isothermal inactivation study, with three water activities and three temperatures, generated data used to evaluate the effectiveness of three secondary models

(response surface, modified Bigelow-type, and a combined-effects model) in accounting for product water activity. The combined-effects model best accounted for water activity when modeling microbial inactivation under low-moisture conditions, when considering goodness-of- fit, phenomenological basis, and model utility.

ACKNOWLEDGMENTS

I would like to take this opportunity to acknowledge a few individuals who have helped me along the way. First and foremost, I would like to thank my advisor, Dr. Bradley Marks, for all of the advice and guidance he has given me over these past two years. Through his example, I have learned to think about problems more critically and it has made me a better engineer.

I would also like to thank Dr. Kirk Dolan and Dr. Elliot Ryser, the remaining members of my master’s committee for their help and guidance. I would especially like to thank Dr. Dolan for taking time to help me with parameter estimation and other modeling topics, and Dr. Ryser for his expertise in microbiology.

I would also like to thank Michael James and Nicole Hall for their help with this project.

I would never have figured out how to obtain the materials I needed without Mike, and Nicole helped me with anything I needed in the lab. Additionally, I would like to thank the other graduate students (Beatriz Mazon, Francisco Garces, Ian Hildebrandt, Pichamon

Limcharoenchat, and Jessica Bruin) in our research group for helping me along the way with questions and providing me with a break when the research, data collection, and writing was becoming difficult. Also, I would not have completed any of these studies without Sarah

Buchholz and Hannah Pichner, the undergraduate workers who stayed late and came back when they didn’t have to.

Finally, I would like to thank my family and all of my friends for supporting me over these last two years. Whenever my frustrations and doubts were getting the best of me, they were the ones who helped me power through. I couldn’t have finished this project without them.

iii

This research was partially supported by funding from the USDA National Institute of

Food and Agriculture NIFA, award 2011-51110-30994.

TABLE OF CONTENTS

LIST OF TABLES ...... vii

LIST OF FIGURES ...... ix

INTRODUCTION ...... 1 1.1 Problem Statement ...... 1 1.2 Goals and Objectives ...... 3

LITERATURE REVIEW ...... 4 2.1 Introduction ...... 4 2.2 Salmonella in Low-Moisture Foods ...... 5 2.2.1 Disease caused by Salmonella ...... 5 2.2.2 Outbreaks ...... 5 2.2.3 Growth and survival in low-moisture foods ...... 7 2.2.4 Desiccation stress mechanisms ...... 8 2.3 Increased Thermal Resistance at Low Water Activity ...... 9 2.4 Water Activity Equilibration and Control ...... 10 2.5 Modeling Microbial Inactivation in Low-Moisture Foods ...... 11 2.5.1 Primary models ...... 11 2.5.2 Secondary models ...... 13 2.5.3 Existing models for water activity effects ...... 14 2.5.4 Model selection and validation ...... 16 2.6 Conclusions ...... 17

EFFECT OF RAPID PRODUCT DESICCATION OR HYDRATION ON THERMAL RESISTANCE OF SALMONELLA ENTERITIDIS PT30 IN WHEAT FLOUR ...... 19 3.1 Objective ...... 19 3.2 Materials and Methods ...... 19 3.2.1 Wheat flour ...... 19 3.2.2 Bacterial strain and inoculation ...... 20 3.2.3 Sample preparation...... 21 3.2.4 Rapid desiccation treatment...... 22 3.2.5 Additional hold time experiment...... 23 3.2.6 Rapid hydration treatment...... 24 3.2.7 Thermal treatment...... 24 3.2.8 Recovery and enumeration...... 25 3.2.9 Statistical analysis...... 25 3.3 Results and Discussion ...... 26 3.3.1 Wheat flour ...... 26 3.3.2 Initial inoculation...... 26 3.3.3 Product water activity ...... 26 3.3.4 Thermal inactivation and D80°C values ...... 27

3.3.4 Variability of rapidly desiccated group...... 31 3.4 Conclusions ...... 32

MODELING THE EFFECT OF TEMPERATURE AND WATER ACTIVITY ON THE THERMAL RESISTANCE OF SALMONELLA ENTERITIDIS PT 30 IN WHEAT FLOUR ... 33 4.1 Objective ...... 33 4.2 Methods and Materials ...... 33 4.2.1 Wheat flour ...... 33 4.2.2 Bacterial strain and inoculation ...... 34 4.2.3 Sample preparation ...... 35 4.2.4 Thermal treatment ...... 35 4.2.5 Recovery and enumeration ...... 36 4.2.6 Primary models ...... 36 4.2.7 Model selection and evaluation ...... 37 4.2.8 Secondary models ...... 38 4.3 Results and Discussion ...... 40 4.3.1 Wheat flour ...... 40 4.3.2 Initial inoculation ...... 40 4.3.3 Comparison of primary models ...... 40 4.3.4 Estimation of parameters for secondary models ...... 42 4.3.5 Comparison of secondary models ...... 45 4.4 Conclusions ...... 48

OVERALL CONCLUSIONS AND RECOMMENDATIONS...... 49 5.1 Effect of Desiccation or Hydration on Thermal Resistance ...... 49 5.2 Modeling the Effects of Temperature and Water Activity on Thermal Resistance ...... 50 5.3 Future Work ...... 51 5.3.1 Cellular mechanisms behind desiccation stress ...... 51 5.3.2 Standardized model selection practices ...... 51 5.3.3 Further validation of the combined-effects model ...... 51

APPENDICES ...... 53 Appendix A – Homogeneity of Inoculation ...... 54 Appendix B – Survivor Data for Rapid Desiccation and Hydration Study ...... 55 Appendix C – Results of ANOVA ...... 59 Appendix D – Photographs of Experimental Apparatus ...... 62 Appendix E – Rapid Desiccation and Hydration System Calibration Data ...... 64 Appendix F – Additional One Hour Hold Data ...... 65 Appendix G – Survivor Data for Factorial Modeling Study ...... 66 Appendix H – MATLAB Code for Log-Linear Model ...... 69 Appendix I – MATLAB Code for Weibull Model ...... 71 Appendix J – MATLAB Code for Secondary Models ...... 73 Appendix K – Surface Plot of the Combined-Effects Model Response ...... 77

REFERENCES ...... 78

LIST OF TABLES

Table 1. Decimal reduction values (D80°C) for all treatments based on linear regression of the survivor curves (Figure 3) from the thermal inactivation treatments...... 28

Table 2. Parameter estimates (± standard error) for the primary models, as well as the root mean square error (RMSE), and the corrected Aikake Information Criterion (AICc) value. Parameters were estimated separately for each data set...... 41

Table 3. Statistical ordinary least squares (OLS) parameter estimates for the response surface model (equation 14)...... 43

Table 4. Statistical ordinary least squares (OLS) parameter estimates for the modified Bigelow- type model (equation 15)...... 43

Table 5. Statistical ordinary least squares (OLS) parameter estimates for the combined-effects model (equation 16)...... 44

Table 6. Statistical ordinary least squares (OLS) results for the secondary models (n = 230). Parameters were globally estimated...... 45

Table 7. Homogeneity test results ...... 54

Table 8. Raw Salmonella survival data I60/F60 ...... 55

Table 9. Raw Salmonella survivor data I30/F30 ...... 56

Table 10. Raw Salmonella survivor data I60/F30 ...... 57

Table 11. Raw Salmonella survivor data I30/F60 ...... 58

Table 12. ANOVA comparison of I60/F60 & I30/F30...... 59

Table 13. ANOVA comparison of I60/F60 & I60/F30...... 59

Table 14. ANOVA comparison of I60/F60 & I30/F60...... 60

Table 15. ANOVA comparison of I30/F30 & I60/F30...... 60

Table 16. ANOVA comparison of I30/F30 & I30/F60...... 60

Table 17. ANOVA comparison of I60/F30 & I30/F60...... 61

Table 18. Desiccation system calibration data ...... 64

vii

Table 19. Hydration system calibration data ...... 64

Table 20. One hour hold data (I60/F30/1H) ...... 65

Table 21. Salmonella survival in wheat flour (aw = 0.30)...... 66

Table 22. Salmonella survival in wheat flour (aw = 0.45)...... 67

Table 23. Salmonella survival in wheat flour (aw = 0.60)...... 68

viii

LIST OF FIGURES

Figure 1. Schematic of the custom-build fluidized-bed drying system for rapid desiccation ...... 23

Figure 2. Schematic of the custom-build fluidized-bed hydrating system for rapid hydration .... 24

Figure 3. Survivor curves for Salmonella Enteritidis PT 30 inoculated into wheat flour during isothermal heating at 80°C. The means of the three replicates for each treatment are shown. .... 27

Figure 4. Thermal inactivation kinetics (80°C) of Salmonella Enteritidis PT 30 inoculated in wheat flour, comparing groups that initially were equilibrated to: (a) 0.60 aw (I60) or (b) 0.30 aw (I30)...... 29

Figure 5. Thermal inactivation kinetics (80°C) of Salmonella Enteritidis PT 30 inoculated in wheat flour, comparing groups with: (a) 0.60 aw (F60) or (b) 0.30 aw (F30) at the time of isothermal heat treatment...... 31

Figure 6. Salmonella D values (min) as a function of wheat flour aw and temperature, predicted by the combined-effects model (equation 16)...... 44

Figure 7. Thermal inactivation of Salmonella inoculated into wheat flour equilibrated to 0.45 aw and treated at 75°C (), 80°C (), or 85°C (). Lines are combined-effects model predictions from the global fit of equation 16 integrated into equation 10...... 46

Figure 8. Thermal inactivation of Salmonella inoculated into wheat flour at 0.60 (), 0.45 (), or 0.30 () aw when treated at 85°C. Lines are combined-effects model predictions from the global fit of equation 16 integrated into equation 10...... 47

Figure 9. Rapid desiccation system ...... 62

Figure 10. Saturation chamber in rapid hydration system ...... 62

Figure 11. Test cell filled with wheat flour that was used in thermal inactivation studies...... 63

Figure 12. Equilibration chamber of the conditioning system ...... 63

Figure 13. Surface response of the D values (min) with respect to aw and temperature (3D)...... 77

INTRODUCTION

1.1 Problem Statement

Recently, low-moisture foods have made headlines because of numerous Salmonella outbreaks and recalls (Li et al., 2014). These incidents have directed increased research attention on the safety of low-moisture foods, which had previously not been considered a food safety risk, because their environments do not support bacterial growth. However, from 2007 – 2012, low-moisture foods accounted for ~20% of all Salmonella outbreaks (CDC, 2014). Products such as black pepper, wheat flour, peanut butter, and whole almonds all have been involved in

Salmonella outbreaks (Gieraltowski et al., 2013; Isaacs et al., 2005; Keady et al., 2004;

McCallum et al., 2013; Sheth et al., 2011). Because low-moisture foods encompass a variety of products, many of which are ingredients in other ready-to-eat foods, such as nuts, food powders, and pastes (e.g., peanut butter), hundreds of finished products could be contaminated, which increases the potential breadth of outbreaks and recalls.

Outbreaks and/or recalls have a large negative impact on the economy. Each year, it is estimated that there are over one million cases of salmonellosis just in the United States (CDC,

2011; Scallan et al., 2011). Further, there are close to 20,000 hospitalizations, which cause ~$51 billion in annual health related costs (Scharff, 2012). Aside from the cost of illness, outbreaks and recalls can negatively impact the economics of the food industry because of potential lawsuits, expensive product recovery, and possible negative public opinion (Elgazzar & Marth,

1992). There is a need to improve food safety processes and systems because of the substantial societal impact.

The bad publicity and large costs associated with recalls and outbreaks have contributed to the development of new guidelines and regulations. In 2011, the FDA Food Safety

Modernization Act (FSMA) was enacted to shift the focus from responding to contamination to actively preventing it (FDA, 2014). In addition to FSMA, the Grocery Manufacturers

Association (GMA) developed guidelines for the control of Salmonella in low-moisture foods

(GMA, 2009b). These guidelines describe common sources of contamination and methods to prevent the spread of Salmonella. The Alliance for Innovative and Operational Excellence

(AIOE) also published a general thermal process validation guidelines document for low- moisture foods (D. G. Anderson & Lucore, 2012). These guidelines are standards the industry may use to validate their low-moisture process and ensure that the process is meeting expected federal rules for preventive controls.

However, simply creating guidelines does not automatically ensure food safety, without the knowledge, data, and tools needed to enable implementation and validation of appropriate processing technologies. For example, modeling microbial inactivation has proven to be quite challenging, because the response of Salmonella is heavily influenced by many product and processing factors. Improved models are needed to sufficiently account for these factors and thereby enable more accurate and robust calculations of process lethality. To achieve this overall goal, some key existing knowledge gaps need to be addressed, regarding the effect of changing water activity on Salmonella thermal resistance.

1.2 Goals and Objectives

Therefore, the overall goal of this thesis project was to better quantify Salmonella thermal resistance in low-moisture powders and to use that knowledge to develop improved inactivation models. The specific objectives were to:

1. Evaluate the effect of rapid product desiccation or hydration on the thermal resistance of

Salmonella Enteritidis PT 30 in wheat flour.

2. Test multiple secondary models for the effect of product (wheat flour) water activity on

Salmonella thermal resistance.

LITERATURE REVIEW

2.1 Introduction

Salmonella is a gram-negative pathogen that is the leading bacterial cause of food poisoning cases in the United States (CDC, 2011). The organism is widely distributed throughout nature, but its primary reservoir is the gastrointestinal tract of humans and animals. Once a sufficient dose of Salmonella has been ingested, salmonellosis can occur in patients, which can cause nausea, vomiting, diarrhea, cramps, and fever (FDA, 2012).

Unfortunately, Salmonella can survive for long periods in low-moisture environments.

Recent recalls of low-moisture foods, such as peanut butter (Medus et al., 2009), wheat flour

(McCallum, et al., 2013), and whole almonds (CDC, 2004), have prompted researchers to study the behavior of Salmonella in low-moisture foods, which had previously been considered safe with little to no validation. Because low-moisture foods encompass a variety of products, the potential pool of foods that could be contaminated is considerable. This large group of products, many of which are used as ingredients in hundreds of ready-to-eat products, greatly expands the types of products that can be implicated in outbreaks and recalls.

Outbreaks and recalls negatively impact the economy because of the rising cost of illness and labor lost as a result of workers taking sick days to recover (Scharff, 2012). In addition, potential lawsuits, lost market share, cost of product return, and negative public opinion are all possible negative outcomes that can be avoided by making low-moisture products safer.

The FDA Food Safety Modernization Act (FSMA) was enacted in 2011 to shift from the culture of being reactive to outbreaks to a culture of actively preventing contamination from occurring by establishing preventative controls (FDA, 2014). By establishing guidelines for the

4 control of Salmonella in low-moisture foods, the industry and regulators can better prevent contamination and protect the general public.

2.2 Salmonella in Low-Moisture Foods

2.2.1 Disease caused by Salmonella

Approximately one in every six Americans gets sick, and 3,000 die each year from foodborne diseases (CDC, 2011). Of these 3,000 deaths, approximately 35% can be attributed to

Salmonella, which makes it an important issue to resolve (He, Guo, Yang, Tortorello, & Zhang,

2011). Foodborne gastroenteritis from Salmonella causes nausea, vomiting, diarrhea, cramps, and fever (FDA, 2012). Symptoms typically last a couple of days and taper off within a week.

For those who possess healthy immune systems, the disease is usually very mild; however, it can be deadly for those with compromised immune systems, such as infants and the elderly.

2.2.2 Outbreaks

Low-moisture foods (those with a water activity, aw, less than 0.60) have recently been implicated in multiple outbreaks and recalls, which have, in turn, focused more research and regulatory attention on the behavior of Salmonella in low-moisture foods. Various product types

(e.g., pastes, large particulates, and powders) across the range of low-moisture foods have all been implicated.

Pastes (e.g., peanut butter) have been responsible for several high profile Salmonella outbreaks. One of the first reported outbreaks from contaminated peanut butter occurred in

Australia in 1996 (Burnett, Gehm, Weissinger, & Beuchat, 2000). Various outbreaks and recalls have occurred since then, but the most prominent occurred in 2008 and 2009. A multistate

5 outbreak of salmonellosis from peanut butter occurred in the United States (Medus, et al., 2009).

Ultimately, 529 cases from 43 states were linked to contaminated peanut butter from a single source.

Low-moisture particulates (e.g., nuts) were previously considered a raw product, meaning there was no required pasteurization of the product before distribution. Almonds were implicated in two outbreaks in 2000 and 2003, which resulted in the recall of almost six million kilograms of raw almonds (CDC, 2004; Isaacs, et al., 2005; Keady, et al., 2004).

Traditionally, food powders were not considered a high risk product in terms of

Salmonella outbreaks, because they typically are ingredients for formulated products that receive appropriate thermal kill steps before consumption (Sperber & North American, 2007). However, food powders (e.g., spices and flavor coatings) have increasingly been added to ready-to-eat foods, meaning there is no subsequent pasteurization process to ensure product safety. In 1993, a nationwide outbreak of salmonellosis in Germany was linked to contaminated paprika and paprika-powdered potato chips (Lehmacher, Bockemuhl, & Aleksic, 1995). This outbreak confirmed that even extremely low numbers of the organism adapted to the low-moisture state were able to cause illness. In 1996, there was an outbreak of foodborne salmonellosis due to powdered infant formula that was contaminated. There were forty-eight cases involving children over a span of six months before health officials in Spain were able to detect the source and recall the formula (Usera et al., 1996). In 2010, over 150 items containing a hydrolyzed vegetable protein (HVP) product were recalled due to the presence of Salmonella (FDA, 2010).

This product was used as an ingredient in a wide variety of items, such as seasoning mix packets, soup mixes, and ready-to-eat meal products. Most recently, there was an outbreak in New

Zealand due to Salmonella-contaminated wheat flour (McCallum, et al., 2013). The increasing

6 number of outbreaks linked to food powders has increased the research attention on the search for more effective intervention methods.

2.2.3 Growth and survival in low-moisture foods

Salmonella has certain requirements in order to grow in an environment, but this project focuses mainly on temperature and aw. Salmonella can grow between 5 and 45°C; with a typical optimal growth temperature of 35°C (Doyle & Mazzotta, 2000). Water activity is one of the major limiting factors to growth. Salmonella is unable to grow when the aw level is below 0.94

(Bell & Kyriakides, 2002). However, even as the aw of a food product drops below 0.94,

Salmonella is still able to survive. Recent studies have shown that while Salmonella is unable to proliferate in low-moisture environments, it is capable of surviving (Hiramatsu, Matsumoto,

Sakae, & Miyazaki, 2005; Janning, Tveld, Notermans, & Kramer, 1994; Juven et al., 1984;

Podolak, Enache, Stone, Black, & Elliott, 2010). Janning et al. (1994) studied the survival of 18 pathogenic bacteria, including Salmonella, in broth media at 0.20 aw. After an initial decrease in colonies, Salmonella stabilized and survived for very long periods (i.e., 248 to 1,351 days with ≤

1 log reduction). It also was observed that Salmonella was more resistant to desiccation than other bacteria.

Salmonella also is able to survive for extended periods in low-moisture pastes (Burnett, et al., 2000; He, et al., 2011; Park, Oh, & Kang, 2008; Shachar & Yaron, 2006). Burnett et al.

(2000) reported that the populations of Salmonella decreased 2.86 to 4.82 log and 4.14 to 4.50 log in different peanut butter and peanut butter spreads during 24 weeks of storage at 5 and 21°C, respectively. Park et al. (2008) reported that Salmonella populations decreased only 0.15 to 0.65 log and 0.34 to 1.29 log in five commercial peanut butters over a 14 day incubation period at 4

7 and 22°C, respectively. While microbial populations decreased during storage, prolonged storage is insufficient to ensure product safety.

In addition, Salmonella is able to survive in large particulates (e.g., almonds) and food powders (Keller, VanDoren, Grasso, & Halik, 2013; Kimber, Kaur, Wang, Danyluk, & Harris,

2012; Uesugi, Danyluk, & Harris, 2006) Uesugi et al. (2006) studied the ability of Salmonella to survive on raw almond surfaces during prolonged storage. They reported no significant reductions during 550 days of storage at -20°C and 4°C. Kimber et al. (2012) confirmed these results when they reported no reduction in populations of Salmonella on almonds stored at -19°C or 4°C. Keller et al. (2013) reported Salmonella survival in black pepper during up to 8 months of typical storage conditions. Obviously, extended storage alone is insufficient to achieve reduced risk of Salmonella survival; a more direct intervention method is needed.

2.2.4 Desiccation stress mechanisms

Relatively little is known about physiological responses to desiccation stress in

Salmonella, but Gruzdev et al. (2011) reported that there is significant overlap with other stress response networks, meaning desiccation stress may activate the thermal stress network.

However, the cellular mechanisms behind the response of Salmonella to desiccation stress are not well known. Based on other stress network responses, Spector and Kenyon (2012) conjectured possible extracellular and intracellular defenses against desiccation stress, but until further studies are completed, these are merely hypotheses.

2.3 Increased Thermal Resistance at Low Water Activity

Low aw is a barrier to growth for many pathogens, including Salmonella (GMA, 2009a).

In addition to surviving desiccation stress, Salmonella exhibits significantly enhanced thermal resistance at low aw (Archer, Jervis, Bird, & Gaze, 1998; Beuchat & Scouten, 2002; Doyle &

Mazzotta, 2000). Goepfert et al. (1970) reported increased heat resistance as media aw decreased. This study was one of the first to make the connection between aw and heat resistance of Salmonella, but the aw range examined (0.87 – 0.99) did not include aw values corresponding to low-moisture foods. Also, the experiments were conducted in a broth culture, which is a different environment than would be found in a typical low-moisture food.

Other studies have been performed in systems more closely resembling those found in low-moisture foods (Archer, et al., 1998; Farakos, Hicks, & Frank, 2014; Jeong, Marks, & Orta-

Ramirez, 2009; Ma et al., 2009; Villa-Rojas et al., 2013). Villa-Rojas et al. (2013) studied the effect of aw on the thermal inactivation of Salmonella in almond kernel flour. Their research showed that increasing product aw decreased Salmonella thermal resistance; however, their study included a fairly high aw range (0.601 – 0.946), which included samples above the low-moisture range. Also, for each aw, the ranges of temperatures tested were different, making it difficult to directly compare the results.

Salmonella thermal resistance in low-moisture powders is still relatively understudied.

Vancauwenberge et al. (1981) inoculated eight Salmonella serotypes onto corn flour containing

10 to 15% moisture and reported longer thermal decimal reduction times (D-values) at lower moisture contents. Farakos et al. (2014) studied the effect of varying NaCl concentrations on the survival kinetics of Salmonella in low aw (0.23 – 0.58) whey protein powder during heating.

Their findings showed that decreasing product aw increased thermal resistance. Archer et al.

(1998) studied the effect of initial sample aw on the thermal decimal reduction times of

Salmonella in wheat flour. D values increased as the aw of the flour decreased over the range of

0.20 to 0.60. However, because the samples were heated in containers with loose lids in a dry oven, sample aw would have changed during heating, but that information was not reported. The relatively low temperatures (57 to 75°C) studied generated very large thermal D values (~800 min), so there remains a need for additional data at higher temperatures related to many typical commercial processes.

2.4 Water Activity Equilibration and Control

When studying aw effects on thermal resistance, prior studies commonly held samples for some period after inoculation (Archer, et al., 1998; Farakos, Hicks, et al., 2014; Jeong, Marks,

Ryser, & Harte, 2012); however, this practice is not standardized and varies widely. Given that equilibration and aw control during the thermal treatment of samples likely impacts the results of such studies, then controlling, measuring, and reporting sample aw is critically important to the interpretation and utility of the resulting inactivation data and models.

Villa-Rojas et al. (2013) did not equilibrate their samples between inoculation and their thermal treatment; however, sample moisture content was assumed constant, because samples were held in sealed containers during heating. By not equilibrating samples, the true effect of aw on Salmonella thermal resistance in low-moisture might not be sufficiently understood.

Archer et al. (1998) inoculated flour samples and then held them in a sealed container at

37°C for 2-3 h, along with silica gel and a saturated solution of lithium chloride, to generate samples with varying aw. They then applied their thermal treatments and grouped data based on

10 sample aw. In other words, sample aw was variable and measured, but not controlled to specific, repeatable target levels.

Farakos et al. (2014) allowed protein powder samples to equilibrate to target aw values in vacuum desiccators before inoculation; however, they did not report the equilibration period or allow an equilibration period after inoculation. They did maintain constant product aw during thermal treatments by vacuum packaging samples before thermal treatment.

Jeong et al. (2012) allowed almond samples a much longer (6-7 days) equilibration period. Samples were placed in a humidity-controlled chamber during this period so that all samples would have uniform aw. In order to fully understand the effect of aw on Salmonella resistance to lethal treatments in low-moisture foods, it is necessary to equilibrate samples before treatments and control aw during the treatment. Additionally, no prior study is known to have reported the effect of equilibration time, or rate of moisture change, on the thermal resistance of

Salmonella in low-moisture foods.

2.5 Modeling Microbial Inactivation in Low-Moisture Foods

2.5.1 Primary models

Modeling the inactivation of Salmonella in low-moisture foods is one tool used to predict the outcomes of thermal processes. Models are an excellent way to optimize existing process parameters, or suggest new ones, and they support risk management options (McKellar & Lu,

2004).

It is well known that microbial inactivation or survival can be modeled using primary models based on first-order kinetics (Peleg, 2006). First-order kinetics are typically used when survival data appear to be log-linear, a isothermal lethal temperature is used, and the number of

11 surviving colonies decreases as a function of time. The following equation is used to represent first-order kinetics, or the log-linear model:

( ⁄ ) ⁄ (1)

where and are the populations (CFU/g) at times t and 0, respectively; t is the time of the isothermal treatment (min); and D is the time (min) required to reduce the microbial population by 90% at a specified temperature (°C) (Peleg, 2006). When the logarithm of the D value is plotted against the corresponding temperatures, the z value, which is the temperature difference needed to decrease (or increase) the D value by a factor of 10, is the negative inverse of the slope

(McKellar & Lu, 2004; Peleg, 2006), which can be calculated as follows:

(2)

However, survival curves are not always log-linear. One approach when this occurs is to use alternative models for the distributions of thermal resistance. The Weibull model is the most widely used and follows the form:

( ⁄ ) ( ⁄ ) (3)

where and are the populations (CFU/g) at times t and 0, respectively; t is the time of the isothermal treatment (min); δ refers to the overall steepness of the survival curve (min); and n describes the general shape of the curve and whether it is linear (n = 1) or nonlinear (n ≠ 1) and has an upward concavity (n < 1) or a downward concavity (n > 1) (Peleg, 2006).

Other primary models have been utilized (Baranyi & Roberts, 1994; Cerf, 1977;

Geeraerd, Herremans, & Van Impe, 2000), but none are as widely used as the log-linear model

12 and the Weibull model. Regardless of the model choice, primary models alone do not account for environmental (product and process) factors, such as product aw.

2.5.2 Secondary models

Secondary models incorporate the response of one or more parameters of a primary model to changes in temperature (T), pH, aw, or other factors (Valdramidis et al., 2006). Models can be built in many ways, but, in theory, the strongest are mathematical expressions built from theoretical bases (McKellar & Lu, 2004). These models can then be extrapolated and adapted to a wide variety of processes, assuming the theory is solid.

There are many approaches to secondary modeling. Common examples include response surface-type, Arrhenius-type, and Bigelow-type models (Valdramidis, et al., 2006). The response surface secondary model is a polynomial model with interaction terms, such as:

(4)

where the significance of the linear model parameters ( ) are tested by a partial F-test

(Neter, Kutner, Wasserman, & Nachtsheim, 1996).

An Arrhenius-type model structure for the effect of aw on thermal inactivation is based on the model of Cerf et al. (1996), in which the effect of aw, T, and pH on the inactivation rate of

Escherichia coli was studied. Because pH will not be considered in the present study, that model can be reduced to the following:

(5)

where k is the first-order reaction rate ( ⁄ ), and , , and are regression parameters.

The final common secondary model incorporating aw is a Bigelow-type structure. This structure is based on the model structure of Gaillard et al. (1998) for describing the combined effects of temperature, pH (omitted here), and aw on the thermal resistance of Bacillus cereus spores, as follows:

( ) ( ) (6)

where Dref is the time (min) needed to achieve a 1 log reduction in the population at T = Tref and

aw = 1; T is the temperature (°C); Tref is the optimized reference temperature (°C); is the aw increment needed to change the D-value by 10-fold at Tref; and is the temperature increment

(°C) needed to change the D-value by 10-fold at the reference aw (which is set at 1.0 in the above equation).

2.5.3 Existing models for water activity effects

As noted above, several secondary models have been reported in studies attempting to incorporate aw into thermal inactivation models (Cerf, et al., 1996; Farakos, Frank, & Schaffner,

2013; Mattick et al., 2001; Neter, et al., 1996; Valdramidis, et al., 2006; Villa-Rojas, et al.,

2013). Such secondary models tend to be specific to the processing parameters described in the study, which can make it difficult to extrapolate or extend that model to different processing conditions. In addition, very few published models are valid for the range of water activities typically present in low-moisture foods. For example, Cerf et al. (1996) used a modified

Arrhenius equation to incorporate aw into the model, but only at levels above 0.95.

Mattick et al. (2001) was one of the first to address the influence of aw on inactivation of

Salmonella at lower aw (0.65 to 0.90) in broths. They reported that a Weibull model with a polynomial secondary model best fit their data generated from broth solutions. However, when

14 they compared the times their model generated with data from actual food products, the model failed to extrapolate accurately, which suggests that product composition and/or structure is an important consideration when modeling bacterial inactivation in low-moisture foods.

Jeong et al. (2009) modified a traditional thermal inactivation model to account for process conditions in moist-air convection non-isothermal heating of almonds. This was accomplished by adding a term to simulate “surface wetness” in a traditional Bigelow-type model based on D values. The model was successful, but did not directly account for product aw.

Studies by Valdramidis et al. (2006), Villa-Rojas et al. (2013), and Farakos et al. (2013) specifically addressed aw influence on inactivation, and used environments more typical to low- moisture foods. Valdramidis et al. (2006) examined three model structures that incorporated temperature and water activity into the secondary model – response surface, Arrhenius, and

Bigelow-type models using data from potato samples (aw = 0.71 – 0.99). The Bigelow-type model fit their data better (R2 = 0.998) than the polynomial and Arrhenius models (R2 = 0.935 and R2 = 0.955, respectively).

Villa-Rojas et al. (2013) examined two primary models and two secondary models – a polynomial relationship and a Bigelow-type relationship, based on data from ground almonds (aw

= 0.601 – 0.946). The Weibull model, with a polynomial secondary model, best fit their data, which was selected based on the coefficient of determination, or R2 value.

Farakos et al. (2013) examined aw influence on inactivation of Salmonella based on experiments from whey powder (aw = 0.19 – 0.54). They developed a polynomial model that best fit their data; however, that was the only secondary model studied. They performed a robust validation procedure with previously published inactivation data from different food powders

(wheat flour, low-fat peanut meal, non-fat dry milk, whey protein, and low-fat cocoa power).

The reported secondary model’s predicted values had a high correlation (R = 0.94) to the observed values from inactivation studies, which supported the validity of their model.

2.5.4 Model selection and validation

The effectiveness of models can be evaluated in a variety of ways. Probably the most common method utilized is the coefficient of determination (R2):

∑ ̅ ̅̅ ̅̅ ̅̅̅̅̅̅̅ (7) ̅̅̅̅̅̅̅̅̅̅̅̅ ∑ ∑ which indicates how well a model fits the observed data points, but which is insufficient as a sole measure for model selection, especially for nonlinear models, because it may overestimate the success of the model (McKellar & Lu, 2004).

In order to better analyze and compare model quality and utility, the root mean squared error (RMSE) and the Corrected Aikake Information Criterion (AICc) are used because they provide additional and important information (Motulsky & Christopoulos, 2004). Models can be examined based on their RMSE, which quantifies accuracy, or goodness-of-fit, of the model predictions:

∑ ( ) √ (8)

where is the measured log reduction (log CFU/g); is the predicted log reduction from the model (log CFU/g); n is the total number of observations, and p is the number of model parameters.

Additionally, the AICc can be calculated, as shown (Motulsky & Christopoulos, 2004):

( ) (9)

16 where, n is the total number of data points, SS is the sum of squares of the residuals, and K is the number of parameters being estimated plus 1. AICc evaluates whether the decrease in the sums of squares of the residuals justifies the addition of parameters to a model. Lower AICc values indicate the more likely correct model for the data (Motulsky & Christopoulos, 2004).

Model validation is rarely reported, but it is critically important to determine whether a model is valid outside of the data used to estimate the model parameters. One approach to validating a model is to use only 2/3 of collected data, selected randomly, to “train” the model, or estimate the parameters. The remaining 1/3 of data is then used to “test” the model to quantitatively analyze the efficacy of the model.

A more common approach is to generate truly independent data from a different product and/or laboratory to quantify the effectiveness of a model. Farakos et al. (2014) validated secondary models by testing a model developed earlier (Farakos, et al., 2013) with Salmonella survival data previously reported for whole wheat flour, peanut meal, dry milk, whey protein, and cocoa powder. Residuals were obtained by subtracting the predicted value from the observed value. The percentage of residuals within the acceptable residual zone (arbitrarily set as -1 log to

0.5 log) was used to evaluate of the model. However, that study only evaluated and validated one secondary model, which was a very basic response surface model. They did not evaluate multiple secondary models to determine which would better predict the effect of aw on microbial inactivation.

2.6 Conclusions

Some key knowledge gaps exist concerning the effect of aw on Salmonella thermal resistance, and the objectives addressed in this work are aimed at those better understanding the

17 effect of moisture and moisture change rate. Better understanding of the effect of aw status and

“history” will assist with future process validations. It will additionally assist in evaluating secondary models to account for product aw in low-moisture foods. There are existing models; however, no prior studies have systematically compared secondary models used to incorporate the effect of aw and temperature, and selected the most likely correct model to be used for validations across many processing platforms and product types.

EFFECT OF RAPID PRODUCT DESICCATION OR HYDRATION ON THERMAL RESISTANCE OF SALMONELLA ENTERITIDIS PT30 IN WHEAT FLOUR

3.1 Objective

The objective of this study was to evaluate the effect of rapid product desiccation or hydration on the thermal resistance of Salmonella Enteritidis PT 30 in wheat flour. Experiments were designed to specifically evaluate the impact of rapid desiccation or hydration on Salmonella thermal resistance by equilibrating samples after inoculation and controlling water activity, aw, during thermal treatments.

3.2 Materials and Methods

3.2.1 Wheat flour

Soft winter wheat organic white flour was obtained in a single batch from Eden Foods

(Clinton Township, MI). Initial aw was measured via an aw meter (Model 3TE, Decagon Devices,

Pullman, WA). Average particle size was determined by performing a particle distribution study, in which Taylor series sieves (#30 through #200) were stacked initially with 100 g of wheat flour. The stack was shaken for 30 min with a motorized sieve shaker (H-4325, Humboldt

Manufacturing, Elgin, IL). The weights of flour remaining in each sieve after the shaking period were then used to calculate the geometric mean particle diameter of the wheat flour.

Negligible Salmonella population in the source wheat flour was confirmed by collecting ten subsamples from the source flour, diluting 10:1 with 0.1% peptone water (BD, Franklin

Lakes, NJ), and plating serial dilutions onto tryptic soy agar (BD) supplemented with 0.6% (w/v) yeast extract (BD) (TSAYE) supplemented with ammonium ferric citrate (0.05%) (Sigma

Aldrich, St. Louis, MO) and sodium thiosulfate (0.03%) (Sigma Aldrich) (modified TSAYE),

19 which differentiated Salmonella from other contaminating microorganisms by the characteristic black precipitate in the center of the colonies. The plates were incubated at 37°C for 48 h. No colonies detected indicated that if Salmonella were present in the source wheat flour, the counts were too low influence the project results and analysis.

3.2.2 Bacterial strain and inoculation

Salmonella enterica servovar Enteritidis phage type 30 (SE PT30) previously was obtained from Dr. Linda Harris (University of California, Davis), and was used in this study because it previously was shown to be thermally resistant in several low aw materials (Anderson et al., 2013; Jeong, et al., 2012; Villa-Rojas, et al., 2013). The culture was maintained at -80°C in tryptic soy broth (BD) supplemented with 20% (vol/vol) glycerol.

Before use, the culture was subjected to two consecutive transfers (24 h each at 37°C) in tryptic soy broth (BD) supplemented with 0.6% (w/v) yeast extract (BD) (TSBYE) and then streaked to plates (150 by 15 mm) of TSAYE to obtain uniform lawns for each of the four groups. After 24 h of incubation at 37°C, the bacterial lawn was harvested in 20 ml of sterile

0.1% peptone water (BD) per plate. The suspension was centrifuged for 15 min at 3,795 × g. The supernatant was discarded, and the remaining pellet was hand mixed into wheat flour (150 g per group) in a sterile plastic bag until pellet was visibly mixed. After hand mixing, the inoculated wheat flour was stomached (Masticator Basic, Neu-Tec Group Inc., Farmingdale, NY) for 3 min to ensure an even distribution. Uniformity of the inoculum distribution was confirmed by plating six subsamples (~1 g) from a 150 g sample after mixing. Target inoculation level was 108

CFU/g.

Because of the risk of small, Salmonella-inoculated flour particles becoming airborne during sample inoculation and preparation, increased safety protocols were necessary. Lab coats, gloves, safety goggles, and face masks were all required personal protective equipment used over the duration of this experiment. Additionally, sample desiccation and hydration treatments

(described below) were conducted in a biosafety cabinet.

3.2.3 Sample preparation.

The inoculated wheat flour (150 g per group) was transferred to an aw conditioning system to adjust the samples to the target aw (Appendix D). The conditioning system consisted of an equilibration chamber (69 cm x 51 cm x 51 cm) and a custom control system, comprised of relative humidity sensors (AM2303, Aosong Electronics Co., Ltd, Guangzhou, China) inside the equilibration chamber, a desiccation column (filled with silica gel), a hydration column (filled with water), solenoid valves, air pumps (Fusion Air Pump 400, JW Pet, Arlington, TX), and a computer-based control system that monitored and controlled the chamber relative humidity within ~2-5%. Samples were conditioned for 4-6 days to allow the entire sample to equilibrate to the target aw, which was confirmed via the aw meter.

The overall treatment design consisted of four groups of samples. Groups were named using the initially equilibrated aw and the aw at the time of the thermal treatment. For example, group I60/F60 was initially equilibrated to 0.60 aw, which was not altered prior to the thermal treatment, and group I60/F30 was initially equilibrated to 0.60 aw, but was rapidly desiccated

(described below) to a final 0.30 aw immediately before thermal treatment. Groups I60/F60,

I30/F30, and I60/F30 were completed from the same inoculation. Group I30/F60 was run at a

21 later date (from the same original batch of flour), but one additional replication of group I60/F60 was run alongside to confirm statistical equivalency to previously run replications.

3.2.4 Rapid desiccation treatment.

After initial equilibration to 0.60 aw, rapidly desiccated samples (I60/F30) were created using a custom-built fluidized-bed drying system (Figure 1) (Appendix D). All treatments were performed in a biosafety cabinet. An air pump (Gast, 1/8 hp, Benton Harbor, MI) drew ambient air (~20°C) through two sequential desiccation columns (L = 38.1 cm; D = 3.81 cm) containing silica gel desiccant (Dry Pak, Encino, CA). Filters with 83.8 micron opening (180 x 180 woven stainless steel mesh, McMaster-Carr, Elmhurst, IL) were installed immediately before and after the sample chamber. Additionally, a 5 μm line filter was installed immediately after the sample chamber, followed by a high-efficiency particulate absorption (HEPA) filter (Filtrete 64703B,

3M, St. Paul, MN) to further prevent the release of Salmonella with the exit air. Desiccated air

(~20% relative humidity) was pumped through the sample chamber (3.0 m/s, measured with a hot-wire anemometer (Model 407123, Extech, Waltham, MA)) to create a fluidized-bed drying system. The system was calibrated to rapidly desiccate flour samples (~1 g) from nominal aw values of 0.60 to 0.30 in ≤ 4 min, confirmed by desiccating six samples and measuring the final aw.

Figure 1. Schematic of the custom-build fluidized-bed drying system for rapid desiccation

The filtration system was tested for efficacy to ensure that it was working as designed in terms of bio-containment. Six plates of modified (differential) TSAYE were placed in the biosafety hood, and six inoculated samples were run through the system. After the six samples were completed, the six plates were incubated at 37°C for 48 h. No Salmonella growth on the plates confirmed the efficacy of bio-containment for the desiccation/hydration system.

3.2.5 Additional hold time experiment.

In order to further evaluate the relatively high variability observed in lethality after the rapid desiccation treatment (described later), the desiccation treatment was repeated as previously described (three replicates), but with one modification. Instead of immediately performing the thermal treatments after desiccation, a 1 h hold time was added between the desiccation treatment and the thermal treatment to evaluate the impact of the additional time on thermal resistance and repeatability (I60/F30/1H).

3.2.6 Rapid hydration treatment.

All samples were treated in a custom-built fluidized-bed hydration system, similar to the desiccation system described above (Figure 2) (Appendix D). All treatments were performed in a biosafety cabinet. Ambient air was pumped (Gast, 1/8 hp, Benton Harbor, MI) through a 7.5 liter glass container filled 2/3 with water. Airstones (porous stone, Petco, San Diego, CA) dispersed the air through the water, thus increasing the humidity of the air, measured to be ~80% relative humidity. The humidified air was pumped (4.5 m/s) through the sample chamber to create a fluidized bed hydration system. The same filtration system was used as in the desiccation treatments to prevent release of Salmonella in the exit air. The system was calibrated to rapidly hydrate flour samples (~1 g) from nominal aw values of 0.30 to 0.60 in 2.5 min, confirmed by hydrating six samples and measuring the final aw.

Figure 2. Schematic of the custom-build fluidized-bed hydrating system for rapid hydration

3.2.7 Thermal treatment.

To obtain thermal death curves for Salmonella, the moisture-equilibrated wheat flour samples were immediately (within ~1 min of the desiccation or hydration treatments) subjected

24 to an isothermal (80°C) inactivation treatment. Aluminum test cells (Chung, Birla, & Tang,

2008) were filled with inoculated and equilibrated flour (0.5 to 0.8 g, 4 mm thick) and immersed in a water bath (Neslab GP-400, Newington, NH) at 80°C (Appendix D). Come-up time was verified using a non-inoculated sample inside a test cell with a K-type thermocouple (20 gauge,

Omega, Stamford, CT) permanently fixed at the center of the sample in the test cell. The come- up time (~68 s) for the sample core to reach within 0.5°C of 80°C was used as time zero for the isothermal treatment. Samples subsequently were removed at equally spaced time intervals, starting with the time zero samples. Once removed from the water bath, the test cells were immediately placed in an ice-water bath to stop the thermal inactivation (T < 40°C in ~20 s).

3.2.8 Recovery and enumeration.

To enumerate Salmonella survivors, thermally treated wheat flour samples were removed from the test cells into sterile plastic bags and diluted 10:1 with 0.1% peptone water (BD).

Appropriate serial dilutions were then plated (0.1 mL) in duplicate onto modified TSAYE. The plates were incubated at 37°C for 48 h. Salmonella counts were converted to log CFU/g. Log reductions were calculated by subtracting survivor counts (log) from the populations (log) at time zero for the respective replicate, which was measured after the come-up time. The limit of detection was 100 CFU/g.

3.2.9 Statistical analysis.

Each treatment group was replicated three times with duplicate samples in each replicate treatment. Linear regression of the ⁄ vs time data was conducted, and the resulting slopes

(-1/D) of the data were compared via analyses of variance (ANOVA), with α = 0.05.

3.3 Results and Discussion

3.3.1 Wheat flour

Initial aw was 0.462, and geometric mean particle size was 144 ± 60 μm. No Salmonella colonies were detected in the source flour after diluting and plating on modified TSAYE.

3.3.2 Initial inoculation.

The average initial Salmonella population in the inoculated samples across all tests was

8.20 ± 0.19 log CFU/g. The standard deviation of the inoculation across six subsamples (~1 g) randomly pulled from a single inoculated sample (150 g) was 0.17 log CFU/g, which confirmed the uniformity of inoculum mixing into the samples (Appendix A). The average population after the thermal come-up time was 6.88 ± 0.77 log CFU/g.

3.3.3 Product water activity

The aw of samples in treatment groups I60/F60 and I30/F30 were measured on the day of the thermal treatment. If the sample was within ± 0.05 of the target water activity, the sample was treated. Any flour samples with aw outside of this tolerance were not used. On the day of thermal treatment, I60/F60 aw was measured (n = 3) to be 0.583 ± 0.003, and I30/F30 aw was measured (n = 3) to be 0.306 ± 0.004.

In preliminary tests, the uniformity and consistency of the desiccation and hydration treatments were confirmed before running any inoculated samples, by testing the aw of six different samples subjected to each of the treatments (Appendix E). These samples were not used for inactivation data, but rather to confirm the repeatability of the desiccation/hydration system.

For I60/F30, samples began at 0.648 ± 0.004 aw and after the desiccation treatment were 0.317 ±

0.012 aw. For I30/F60, samples began at 0.306 ± 0.002 aw and after the hydration treatment were

0.602 ± 0.018 aw. The means of the sample aw at the time of thermal treatment for groups

I60/F60 and I30/F60 (0.583 ± 0.003 and 0.602 ± 0.018, respectively) were not significantly different (as shown by a t-test), nor were the means of the aw at the time of thermal treatment for groups I30/F30 and I60/F30 (0.306 ± 0.004 and 0.317 ± 0.012, respectively). Sample aw within ±

0.05 of the target was assumed to have a negligible effect on the thermal resistance (given the final results described below), so that Daw=0.306 would be expected to be Daw=0.317, and Daw=0.583 would be expected to be Daw=0.602.

3.3.4 Thermal inactivation and D80°C values

Survivor curves (Figure 3) were calculated using all of the data from the three replicates for each sample (although the figure shows only the means of the three replicates).

1 I60/F60

0 I30/F30

I60/F30 -1 I30/F60

[log [log CFU/g] -2

-3 Log N/N Log -4

-5 0 200 400 600 800 1000 1200 Time [s]

Figure 3. Survivor curves for Salmonella Enteritidis PT 30 inoculated into wheat flour during isothermal heating at 80°C. The means of the three replicates for each treatment are shown.

D80°C values were calculated by taking the inverse of the slopes of each replicate, which were generated from the thermal inactivation treatments of the four groups, then averaging the slopes of the three replicates of one treatment to generate one D80°C value per treatment ± standard deviation, (Table 1) (raw data in Appendix B). The slopes (-1/D) for treatments I60/F60 and I30/F60 were statistically equivalent (P > 0.05), and the slopes for treatments I30/F30 and

I60/F30 were statistically equivalent (P > 0.05); however, the D80°C value for I60/F30 was approximately five times larger than the D80°C value for I30/F60 (Appendix C). The common factor within the groups that were not significantly different from each other was aw at the time of thermal treatment, regardless of the initially equilibrated aw prior to desiccation or hydration.

This result suggests that the aw state at the time of treatment controls the thermal inactivation of

Salmonella, regardless of the moisture “history” prior to thermal treatment. In prior work on the heat resistance of Salmonella Enteritidis PT30 in almond kernel flour, Villa-Rojas et al. (2013) reported a D80°C value of 1.63 min for 0.601 aw, which is very similar to the value found in this study. However, they did not examine aw values below 0.60.

Table 1. Decimal reduction values (D80°C) for all treatments based on linear regression of the survivor curves (Figure 3) from the thermal inactivation treatments. D80°C value [min] Treatment Group (mean ± SD)a R2 I60/F60 1.33 ± 0.15 A 0.496 I30/F30 7.32 ± 2.07 B 0.719 I60/F30 5.73 ± 3.10 B 0.405 I30/F60 1.15 ± 0.09 A 0.726 I60/F30/1H 5.98 ± 2.34 B 0.568 a Different letters within the column indicate that means are significantly different (P < 0.05).

Figure 4 shows the treatment groups with stationary water activities (I60/F60 and

I30/F30) compared to the treatment groups that were initially equilibrated to the same aw. Group

I60/F60 and I60/F30 (Fig 4a) both were initially equilibrated to 0.6 aw and are significantly different (P < 0.05). Group I30/F30 and I30/F60, which were initially equilibrated to 0.3 aw, were significantly different (P < 0.05). Figures 4a and 4b show the significantly different slopes between the groups. These data suggest that the treatment had a significant effect on the thermal resistance of Salmonella.

1 (a)

0 I60/F60

-1 I60/F30

[log [log CFU/g] -2

-3 Log N/N Log -4

-5 0 500 1000 Time [s]

1 (b) I30/F30 0 I30/F60 -1

[log [log CFU/g] -2

N/N -3 Log Log -4

-5 0 500 1000 Time [s]

Figure 5 shows the treatment groups with stationary water activities compared to the treatment groups with similar water activities at the time of thermal treatment. Group I60/F60 and I30/F60 (Fig 5a) were not significantly different (P > 0.05). Similarly, Group I30/F30 and

I60/F30 (Fig 5b) were not significantly different (P > 0.05). These results also can be seen visually as their slopes are very close together. They also suggest that aw “history” does not affect Salmonella thermal resistance, which is controlled by the aw at the time of thermal treatment. It also suggests that the time required for Salmonella thermal resistance to respond to new water activities is effectively negligible.

1 (a)

-1 I60/F60

[log [log CFU/g] -2 I30/F60

N/N -3 Log Log -4

-5 0 500 1000 Time [s]

1 (b) I30/F30 0 I60/F30 -1

[log [log CFU/g] -2

N/N -3 Log Log -4

-5 0 500 1000 Time [s]

3.3.4 Variability of rapidly desiccated group

Some of the treatment groups had relatively high variability among replicates in terms of log reductions, with treatment I60/F30 (rapid desiccation) showing the highest variability. Little is known about the physiological response to desiccation stress, but this variability could be partially due to cell components within Salmonella that protect against desiccation, and

31 population variability in this response depending on the individual cell (Spector & Kenyon,

2012). However, because the focus of this study was to quantify the effects of rapid aw change, the specific cellular mechanisms involved were not examined.

As described in the methods, an additional treatment with a 1 h hold time between the desiccation treatment and the thermal treatment was run to observe any impacts on the response period or variability in the results (raw data in Appendix F). No significant difference was observed in the slope between I60/F30 and I60/F30/1H, but the standard deviation within the replication decreased from 1.20 (immediate) to 0.45 (1 h wait). Therefore, variability among replicates was reduced by the 1 h hold, but the mean result was unchanged. This may suggest

“equilibration” of the response across the population, but there was no net effect on the thermal resistance.

3.4 Conclusions

These results reinforce the premise that Salmonella thermal resistance increases with decreasing aw at the time of thermal treatment, supporting findings from previous studies

(Archer, et al., 1998; Janning, et al., 1994). However, prior to this study, the response period was not tested or reported, and the present study is the first known to demonstrate the effect of moisture change rate on that resistance. Knowledge regarding the effect of both the extent and rate of material desiccation or hydration on Salmonella thermal resistance in low-moisture foods in critically important in validating pasteurization processes or any “kill step” in commercial processes. Future secondary inactivation models, and their application to process validation, clearly need to account for dynamic moisture during processing.

MODELING THE EFFECT OF TEMPERATURE AND WATER ACTIVITY ON THE THERMAL RESISTANCE OF SALMONELLA ENTERITIDIS PT 30 IN WHEAT FLOUR

4.1 Objective

The objective of this study was to test multiple secondary models for the effect of product

(wheat flour) water activity, aw, on Salmonella thermal resistance. Two primary models and three secondary models were evaluated to determine which best described the inactivation of

Salmonella in wheat flour during isothermal treatment.

4.2 Methods and Materials

4.2.1 Wheat flour

Soft winter wheat organic white flour was obtained in a single batch from Eden Foods

(Clinton Township, MI). Initial aw was measured via an aw meter (Model 3TE, Decagon Devices,

Manufacturing, Elgin, IL). The weights of flour remaining in each sieve after the shaking period were then used to calculate the geometric mean of the wheat flour.

Negligible Salmonella population in the source wheat flour was confirmed by collecting ten subsamples from the source flour, diluting 10:1 with 0.1% peptone water (BD, Franklin

Lakes, NJ), and plating serial dilutions onto tryptic soy agar (BD) supplemented with 0.6% (w/v) yeast extract (BD) (TSAYE) supplemented with ammonium ferric citrate (0.05%) (Sigma

Aldrich, St. Louis, MO) and sodium thiosulfate (0.03%) (Sigma Aldrich) (modified TSAYE), which differentiated Salmonella from other contaminating microorganisms by the characteristic

33 black precipitate in the center of the colonies. The plates were incubated at 37°C for 48 h. No colonies detected indicated that if Salmonella were present in the source wheat flour, the counts were too low to affect the project results and analyses.

4.2.2 Bacterial strain and inoculation

Salmonella enterica servovar Enteritidis phage type 30 (SE PT30)previously was obtained from Dr. Linda Harris (University of California, Davis), and was used in this study because it previously was shown to be thermally resistant in several low aw materials (Anderson, et al., 2013; Jeong, et al., 2012; Villa-Rojas, et al., 2013). The culture was maintained at -80°C in tryptic soy broth (BD) supplemented with 20% (vol/vol) glycerol.

Before use, the culture was subjected to two consecutive transfers (24 h each at 37°C) in tryptic soy broth (BD) supplemented with 0.6% (w/v) yeast extract (BD) (TSBYE) and then streaked to plates (150 by 15 mm) of TSAYE to obtain uniform lawns for each replication. After incubation (24 h, 37°C), the bacterial lawn was harvested in 20 ml of sterile 0.1% peptone water

(BD) per plate. The suspension was centrifuged for 15 min at 2,988 × g. The supernatant was discarded, and the remaining pellet was hand mixed into wheat flour (25 g per replication) in a sterile plastic bag until the pellet was visibly mixed. After hand mixing, the inoculated wheat flour was stomached (Masticator Basic, Neu-Tec Group Inc., Farmingdale, NY) for 3 min to ensure an even distribution. Uniformity of the inoculum distribution was confirmed by plating six random subsamples (~1 g) from a 25 g sample after mixing. Target inoculation level was 109

CFU/g.

Because of the risk of small, Salmonella-inoculated flour particles becoming airborne during sample inoculation and preparation, increased safety protocols were necessary. Lab

34 coats, gloves, safety goggles, and face masks were all required personal protective equipment used for the duration of this experiment.

4.2.3 Sample preparation

The inoculated wheat flour (25 g per replication) was transferred to an aw conditioning system to adjust the samples to target aw levels (Appendix D). The conditioning system consisted of an equilibration chamber (69 cm x 51 cm x 51 cm) and a custom control system, comprised of relative humidity sensors (AM2303, Aosong Electronics Co., Ltd, Guangzhou,

China) inside the equilibration chamber, a desiccation column (filled with silica gel), a hydration column (filled with water), solenoid valves, air pumps (Fusion Air Pump 400, JW Pet, Arlington,

TX), and a computer-based control system that monitored and controlled the chamber relative humidity within ~2-5%. Prior to thermal treatment, samples were conditioned for 4-6 days at the target humidity (30, 45, or 60% relative humidity) to allow the entire sample to equilibrate to the target aw, which was subsequently confirmed via the aw meter for each target aw.

4.2.4 Thermal treatment

A full factorial experiment was used with three constant inactivation temperatures (75,

80, and 85°C) and three constant water activities (nominally 0.30, 0.45, and 0.60), with all experiments conducted in triplicate.

To obtain thermal death curves for Salmonella, the moisture-equilibrated wheat flour samples were subjected to an isothermal inactivation treatment. Aluminum test cells (Chung, et al., 2008) were filled with inoculated and equilibrated flour (0.5 to 0.8 g, 4 mm thick) and immersed in a water bath (Neslab GP-400, Newington, NH) at 75, 80, or 85°C (Appendix D).

Come-up time was verified using a non-inoculated sample inside a test cell with a K-type thermocouple (20 gauge, Omega, Stamford, CT) permanently fixed at the geometric center of the flour sample in the test cell. The come-up time (64-70 s) for the sample core to reach within

0.5°C of the target temperature was used as time zero for the isothermal treatment. Samples were subsequently removed at uniform time intervals, starting with the time zero samples. Once removed from the water bath, the test cells were immediately placed in an ice-water bath to stop the thermal inactivation (T < 40°C in ~20 s).

4.2.5 Recovery and enumeration

To enumerate Salmonella survivors, thermally treated wheat flour samples were removed from the test cells into sterile plastic bags and diluted 10:1 with 0.1% peptone water (BD).

Appropriate serial dilutions then were plated (0.1 mL) in duplicate onto modified TSAYE. The plates were incubated (37°C for 48 h), and Salmonella colonies were enumerated and populations converted to log CFU per gram. Log reductions were calculated by subtracting survivor counts (log) from the population (log) at time zero for the respective replicate, which was measured after the come-up time. The limit of detection was 100 CFU/g.

4.2.6 Primary models

The two most widely used primary models in bacterial inactivation are the first-order kinetic model and the Weibull-type model. The first-order kinetic, or log-linear, model (Peleg,

2006) was:

( ⁄ ) ⁄ (10)

36 where and are the populations (CFU/g) at times and 0 respectively, is the time of the isothermal treatment (min) after come-up, and is the time (min) required to reduce the microbial population by 10-fold at a specified temperature (°C).

The Weibull model (Peleg, 2006) was:

( ⁄ ) ( ⁄ ) (11)

where δ refers to the overall steepness of the survival curve (min), and n describes the general shape of the curve and describes whether it is linear (n = 1) or nonlinear (n ≠ 1) with a decreasing

(n < 1) or increasing (n > 1) inactivation rate with time.

Parameters were estimated for each of the nine data sets separately using ordinary least squares (OLS) minimization in MATLAB (version 2014). The parameters were estimated for each of the nine data sets by performing OLS with nlinfit (nonlinear regression routine in the statistical toolbox) (Appendices H and I).

4.2.7 Model selection and evaluation

Models were fit to the data, and the goodness-of-fit for each candidate model was quantified by the root mean square error (RMSE) (Motulsky & Christopoulos, 2004):

∑ ( ) √ (12)

where is the measured log reduction (log CFU/g), is the predicted log reduction from the model (log CFU/g), n is the total number of observations, and p is the number of model parameters.

Additionally, the Corrected Aikake Information Criterion (AICc) was calculated for each model (Motulsky & Christopoulos, 2004):

( ) (13)

where, n is the total number of data points, SS is the sum of squares of the residuals, and K is the number of parameters being estimated plus 1. AICc evaluates whether the decrease in the sum of squares of the residuals justifies the addition of parameters to a model. Lower AICc values indicate the model more likely to be correct for the data.

Other statistical indices used to compare the models were the relative error of each parameter, confidence intervals, correlations among the parameters, and the residuals.

4.2.8 Secondary models

Based on the AICc (results shown below), the log-linear model was the more likely correct primary model. Therefore, three secondary models were evaluated for the influence of aw and temperature on D values in the log-linear model. The first secondary model was a second- order response surface model.

(14)

The statistical significance of each linear model parameter ( ) was tested by a partial F- test (Neter, et al., 1996; Valdramidis, et al., 2006).

The second secondary model evaluated was a modified version of a Bigelow-type relationship (Gaillard, et al., 1998).

(15)

where Dref is the time (min) needed to achieve a 1 log reduction in the population at Tref and aw =

1, T is the temperature (°C), Tref is the optimized reference temperature (°C), is the aw

38 increment needed to change the D-value by 10-fold at the Tref, and is the temperature increment needed to change the D-value by 10-fold (°C) at the reference aw.

The final secondary model was a combined-effects model that was built from assumptions based on preliminary data analyses. Preliminary data suggested that the effect of water activity on the D value is linear, while the effect of temperature is non-linear. With these observations, a modified model was created by adding a term to account for the linear effect of water activity to the Bigelow model (Bigelow & Esty, 1920), yielding the following model:

(16) [ ] where Dref is the time (min) needed to achieve a 1 log reduction in the population at Tref and aw =

0.60, which was chosen because it is the upper limit of the definition of low-moisture foods, is the resistance added due to a change in aw, T is the temperature (°C);, Tref is the reference temperature (°C), and is the temperature increment needed to reduce the D-value by 10-fold

(°C).

Dolan et al. (2013) reported that the optimal reference temperature should be found for each data set in order to minimize the correlation between parameters. Additionally, relative error is minimized by minimizing the correlation coefficient. The optimal reference temperature was found for each secondary model by varying Tref over a range of temperatures when estimating the parameters, and plotting the temperature range vs. the correlation coefficient. The temperature at which the correlation coefficient is ~0 is the model’s optimal reference temperature.

All parameters for each secondary model were estimated globally using ordinary least squares (OLS) minimization with nlinfit (nonlinear regression routine in the statistical toolbox)

39 in MATLAB (version 2014) for the combined model of equation 10 with either equation 14, 15, or 16 (Appendix J).

4.3 Results and Discussion

4.3.1 Wheat flour

Initial aw was 0.462, and geometric mean particle size was 144 ± 60 μm. No Salmonella was detected in the source flour.

4.3.2 Initial inoculation

The standard deviation of the inoculation across six subsamples (~1 g) randomly pulled from a single inoculated sample (25 g) was 0.17 log CFU/g, which confirmed the uniformity of the inoculum mixing into the samples (Appendix A). The mean initial Salmonella population ± std. deviation across all inoculated samples tests was 9.26 ± 0.25 log CFU/g. The mean

Salmonella population ± std. deviation of all inoculated samples after the thermal come-up time was 8.75 ± 0.60 log CFU/g, which, within each replication, was treated as the time zero population. The difference between the initial inoculation population and the time zero population was significantly different (P < 0.05), which is why it was important to use the time zero population in the subsequent analyses of the isothermal inactivation curves.

4.3.3 Comparison of primary models

Both primary models, the log-linear model and Weibull model, were fit to each of the nine data sets (T, aw) according to equations 10 and 11 (Table 2) (data shown in Appendix G).

The Weibull model showed very slight tailing behavior; however, the log-linear model was the

40 more likely correct model for all of the cases, based on the lower AICc values. Villa-Rojas et al.

(2013) reported that the Weibull model was the better predictive model for isothermal inactivation of Salmonella in ground almonds; however, that statement was based solely on the

R2 values.

For the log-linear model, Villa-Rojas et al. (2013) reported a D80°C value of 1.63 min for

0.601 aw, which is very similar to the D80°C value of 1.27 min for 0.60 aw found in this study.

However, they did not examine aw values below 0.60. Farakos et al. (2014) studied Salmonella in protein powders at lower aw, but they did not report D values, as they utilized a Weibull model.

However, their results indicated a time to 1 log reduction of 6.1 min for 0.58 aw and 80°C, which is larger than the comparable value from this study at 0.6 aw. The differences might be attributed to compositional differences and the fairly large standard errors in both studies.

Log-linear model Weibull Model

RMSE RMSE T δ a D value [min] [log AIC a n [log AIC a w c [min] c [°C] CFU/g] CFU/g] 0.310 75 14.53 ± 0.68 0.3594 -47.65 13.37 ± 1.58 0.89 ± 0.13 0.3619 -45.76  0.005 80 10.27 ± 0.65 0.6090 -23.30 8.62 ± 1.88 0.83 ± 0.17 0.6106 -21.68 85 5.05 ± 0.18 0.3563 -52.24 4.58 ± 0.53 0.90 ± 0.10 0.3573 -50.62

0.427 75 9.97 ± 0.46 0.8091 -7.96 8.08 ± 2.00 0.87 ± 0.13 0.8108 -6.36  0.003 80 5.51 ± 0.22 0.5431 -29.48 4.97 ± 0.84 0.92 ± 0.12 0.5497 -27.34 85 2.11 ± 0.09 0.6652 -16.02 1.59 ± 0.34 0.83 ± 0.11 0.6451 -15.93

0.582 75 2.76 ± 0.19 0.9972 3.62 2.25 ± 0.74 0.88 ± 0.16 1.0107 5.89  0.003 80 1.27 ± 0.06 0.4305 -42.03 1.10 ± 0.16 0.86 ± 0.12 0.4296 -40.66 85 0.65 ± 0.04 0.7035 -15.51 0.63 ± 0.13 0.97 ± 0.17 0.7170 -13.00 a The lower AICc value indicates the model more likely to be correct. Comparisons can only be made horizontally.

4.3.4 Estimation of parameters for secondary models

The parameters of each secondary model (equations 14-16) were estimated globally for all aw and temperatures simultaneously (Table 3-5). The optimal reference temperatures were estimated for the modified Bigelow-type model and the combined-effects model per the method described in Dolan et al. (2013). However, to more easily compare the models, all parameters were estimated with the reference temperature set to 80°C, which was within 4°C of the optimal values, and was the midpoint of the test range.

The response surface model had a relatively high RMSE value (1.59 log CFU/g), and the relative errors for the parameters were extremely high, ranging from 0.71% to 183% (Table 3).

In addition, the confidence intervals indicated that only three of the six parameters were significant. Further, no parameters containing temperature were significant, which indicates an inappropriate model, because temperature is known to affect Salmonella thermal resistance.

There also were high correlations among the parameters (6 out of 15 greater than 0.95), and the residuals contained additive errors and were not normally distributed. These results indicate that the response surface model was not a good secondary model for this application.

The modified Bigelow-type had a lower RMSE (0.745 log CFU/g), and the parameter relative errors were much lower, ranging from 3.15 to 12.1% (Table 4). Examination of the confidence intervals showed that all parameters were significant; however, Dref and zaw were correlated, which could affect the ability to accurately estimate the parameters. The residuals did not contain additive errors and were normally distributed; however, the mean of the residuals was not zero (i.e., -0.14 log CFU/g).

The combined-effects model had the lowest RMSE (0.668), and the parameter relative errors were the lowest for all models, ranging from 2.93 to 3.05% (Table 5). Examination of the

42 confidence intervals showed that all parameters were significant. Further, none of the parameters were correlated, meaning the parameters could be accurately estimated. The residuals did not contain additive errors, were normally distributed, and the mean of the residuals was approximately zero (i.e., -0.05 log CFU/g). The residuals did not appear to be correlated; however, upon testing, they were found to be slightly correlated. These results indicate the combined-effects model performed best of the three secondary models in this application.

Table 3. Statistical ordinary least squares (OLS) parameter estimates for the response surface model (equation 14).

Tref [°C] N/A SSE 566 MSE 2.53 RMSE 1.59 Relative Parameter Estimatea Std. Error 95% Lower CI 95% Upper CI Error

[min] -815 * 212 -1570 -64.3 -26.0% [min/°C] -4.26 5.08 -42.3 33.8 -119%

[min/(°C)2] 0.02 0.03 -0.19 0.23 183% [min] 5760 * 100 1260 10300 1.74% [min] -7900 * 56.4 -15300 -495 -0.71% [min/°C] 1.73 1.67 -12.6 16.1 96.2% aAsterisks indicate which parameters were significant (P < 0.05). Parameter estimates were generated assuming all parameters were significant. Parameters were not re-estimated using only significant parameters.

Table 4. Statistical ordinary least squares (OLS) parameter estimates for the modified Bigelow- type model (equation 15).

Tref [°C] 80.0 SSE 126 MSE 0.556 RMSE 0.745 95% 95% Relative Parameter Estimate Std. Error Lower CI Upper CI Error [min] 0.08 0.01 0.06 0.10 12.1% [°C] 15.2 0.48 14.3 16.2 3.17%

0.32 0.01 0.30 0.34 3.15% 43

Table 5. Statistical ordinary least squares (OLS) parameter estimates for the combined-effects model (equation 16).

Tref [°C] 80.0 SSE 101 MSE 0.446 RMSE 0.668 95% Lower 95% Upper Parameter Estimate Std. Error Relative Error CI CI [s] 1.32 0.04 1.24 1.40 3.05%

[s] 25.3 0.74 23.8 26.7 2.93% [°C] 16.3 0.49 15.3 17.2 3.00%

To further evaluate the combined-effects model, the parameter estimates were used to generate values for the range of aw and temperatures used in this study. The predicted D values were then plotted with respect to aw and temperature (Figure 6) (Appendix K).

0.60 1 2

0.55

4 0.50 6 3

0.45 8 Water ActivityWater 10 0.40 12 5 7 14 9 0.35 11 16 13 15

0.30 75 76 77 78 79 80 81 82 83 84 85 Temperature [C]

Figure 6. Salmonella D values (min) as a function of wheat flour aw and temperature, predicted by the combined-effects model (equation 16).

4.3.5 Comparison of secondary models

Typically, the best model is selected based on the lowest AICc value; therefore, the combined-effects model was selected as the most effective secondary model (Table 6). The surface response model was extremely data and application-intensive because of the number of parameters in the model, and could not be used effectively outside of the range of aw put into the model when estimating the parameters. The modified Bigelow-type model, which is a widely accepted secondary model, predicted the inactivation relatively well (RMSE = 0.745); however, the combined-effects model had a better goodness-of-fit (RMSE = 0.668), and was more likely to be correct based on lower AICc. Further, the parameters of the combined-effects model have a degree of phenomenological meaning to the data set and are not simply coefficients to a model.

Table 6. Statistical ordinary least squares (OLS) results for the secondary models (n = 230). Parameters were globally estimated. SSE # RMSE Optimum T a Model [(log ref AIC Parameters [log CFU/g] [°C] c CFU/g)2] Response Surface 6 566 1.59 N/A 222 Modified Bigelow 3 126 0.745 77.4 -130 Combined-Effects 3 101 0.668 76.5 -181 a Optimum Tref was set to 80°C for parameter estimation, but these values are the optimal Tref for each model. The Response surface model did not explicitly have a Tref, as it was accounted for in the parameter β0.

The combined-effects model was selected as the most comprehensive secondary model because it effectively accounted for temperature and aw. At a specific aw, the combined-effects model accurately discriminated between differing temperature (Figure 7), and at a set temperature, the combined-effects model accurately discriminated between differing aw (Figure

8).

-1

-2

-3

[log CFU/g] 0

-4 Log N/N Log -5 80°C 75°C -6 85°C

-7 0 10 20 30 40 50 60 Time [min]

-1

-2

[Log CFU/g] [Log 0

-3 Log N/N Log -4 aw = 0.30

-5 aw = 0.60 aw = 0.45

-6 0 2 4 6 8 10 12 14 16 Time [min] Figure 8. Thermal inactivation of Salmonella inoculated into wheat flour at 0.60 (), 0.45 (), or 0.30 () aw when treated at 85°C. Lines are combined-effects model predictions from the global fit of equation 16 integrated into equation 10.

Valdramidis et al. (2006) determined that the Bigelow-type model better fit inactivation data generated from Listeria monocytogenes in potato slices (aw = 0.71 to 0.99) when compared with a polynomial and Arrhenius structure, based on R2 values; however, we cannot directly compare these results, because the product, organism, and environment were quite different.

Nevertheless, the results from this study support those from Valdramidis et al. indicating that the

Bigelow-type model is more likely correct than the response surface model. Farakos et al. (2013) evaluated primary and secondary models based on survival data generated from Salmonella in protein powder (aw = 0.19 to 0.54). They determined that the Weibull model better described the survival data, when compared to the log-linear model, based on adjusted R2 and RMSE. They then evaluated a polynomial secondary model integrated into the Weibull model. They subsequently validated that model against independent data (Farakos, Schaffner, et al., 2014), and indicated that their model was fail-safe, with 55% of the prediction residuals being within a

47 pre-defined acceptable range (-1 to 0.5 log). However, this result was only slightly better than the standard Bigelow model (Bigelow & Esty, 1920), which was fail-safe with 52% of the residuals.

4.4 Conclusions

For the data in this study, the combined-effects model performed better than the modified

Bigelow-type model in predicting thermal inactivation as a function of aw and temperature. The combined-effects secondary model also yielded improved outcomes by accounting for aw when modeling microbial inactivation, as compared to a traditional Bigelow model (Bigelow & Esty,

1920). The standard Bigelow model (equation 2) integrated into the log-linear model (equation

10) yielded a RMSE of 1.45 log CFU/g. When the combined-effects model (equation 16), which considers temperature and aw, was integrated into the log-linear model (equation 10), the RMSE was reduced to 0.668 log CFU/g.

In future studies, this model form should be validated with data from other products and treatments; however, it appears to have shown promise for wheat flour. One limitation of the combined-effects model, as presented, is that extrapolation must be avoided, because the model

(with the reported parameters) predicts D values less than zero for aw greater than ~0.65, which is a severe limitation to broader applications. Current tests are extending the data set to higher aw values to estimate updated model parameters that will enable more robust application to a broader range of conditions. Clearly, aw is an important factor affecting thermal inactivation of

Salmonella in low-moisture products, and should be appropriately included in thermal inactivation models for these types of systems.

OVERALL CONCLUSIONS AND RECOMMENDATIONS

5.1 Effect of Desiccation or Hydration on Thermal Resistance

The mechanisms behind desiccation stress and the impact on subsequent thermal resistance are not well known (Spector & Kenyon, 2012). Despite not understanding the cellular mechanisms, it is clear that water activity “history” seems to have little effect on Salmonella thermal resistance.

The effect of desiccation or hydration rate was determined to have little effect on thermal resistance; instead, the water activity, aw, at the time of thermal treatment was determined to control Salmonella thermal resistance, regardless of the prior water activity history. These results reinforce findings from previous studies (Archer, et al., 1998; Janning, et al., 1994), in which decreased aw increased Salmonella thermal resistance. However, the impact of moisture change rate on the resistance had not been previously examined. Understanding both the extent and rate of material desiccation or hydration on Salmonella thermal resistance in low-moisture foods is critically important in validating pasteurization processes or any “kill step” in commercial processes, which typically consist of both non-isothermal and non-iso-moisture conditions (e.g., drying, baking, toasting, or roasting).

During many commercial processes, product water activity fluctuates with changing production environments. Process validations must consider the dynamic nature of product water activity at the time of thermal processing in order to accurately estimate the process lethality.

5.2 Modeling the Effects of Temperature and Water Activity on Thermal Resistance

The modeling study examined the effectiveness of two primary models. After both the log-linear and Weibull models were fit to the data, AICc values showed that the log-linear model better represented the data than did the Weibull model. The apparent linearity of the data also was an indication that the data would be better represented with the log-linear model. For these reasons, along with the inherent simplicity of the model, all secondary models were developed as extensions of the log-linear primary model.

For the data in this study, the combined-effects model performed better than the modified

Bigelow-type model in predicting thermal inactivation as a function of aw and temperature. The combined-effects secondary model was a significant improvement with respect to accounting for aw when modeling microbial inactivation. When the standard Bigelow model (equation 2) was integrated into the log-linear model, the RMSE was 1.45. When the combined-effects model

(equation 16), which accounts for aw, was integrated into the log-linear model, the RMSE decreased to 0.668. Further, the parameters of the combined effects model have phenomenological meaning and are not simply coefficients to a model. Clearly, aw is an important factor affecting thermal inactivation of Salmonella in low-moisture products, and should be appropriately included in thermal inactivation models for these types of systems.

Although these results are slightly different from previous studies (Valdramidis, et al.,

2006; Villa-Rojas, et al., 2013), these prior studies based model selection almost exclusively on the coefficient of determination, or the R2 value, which is an insufficient sole measure for evaluating nonlinear models, such as the Weibull model (McKellar & Lu, 2004). More robust selection statistics, such as the root mean square error, and/or the corrected Aikake Information

Criterion should be used.

Accounting for the product environment is an important part of modeling microbial inactivation. The response surface model and the modified Bigelow-type models studied in this work have previously been applied (Farakos, et al., 2013; Valdramidis, et al., 2006); however, the combined-effects model is a novel effort to better account for the effect of water activity when modeling microbial inactivation for low-moisture conditions.

5.3 Future Work

5.3.1 Cellular mechanisms behind desiccation stress

There is a need for further understanding the cellular mechanisms behind desiccation stress. Deep understanding of these mechanisms was not the purpose of this study, but based on the general responses reported in the present work, future efforts could focus on elucidating the underlying mechanisms, which would in turn help improve future models.

5.3.2 Standardized model selection practices

There is a need for standardized model validation practices when modeling microbial inactivation. Many studies use only the coefficient of determination (R2) as a selection criterion when comparing linear and nonlinear models, which may overestimate the success of the model, leading to poor model selection (McKellar & Lu, 2004).

5.3.3 Further validation of the combined-effects model

The combined-effects model shows promise in the endeavor to incorporate the effect of water activity into microbial inactivation models. However, the model was only confirmed for the process and product conditions in which the data were generated. A future study should

51 validate the combined-effects model using different low-moisture products (e.g., pastes, particulates, powders), different product compositions (e.g., protein, fat), different Salmonella strains, and, most importantly, non-isothermal and non-iso-moisture conditions.

APPENDICES

Appendix A – Homogeneity of Inoculation

In order to ensure the homogeneity of the inoculation methods, six samples (~1 g) were randomly pulled from an inoculated 25 g sample. This test confirmed the uniformity of the inoculum mixing into the flour samples.

Table 7. Homogeneity test results

Water Activity Sample Log CFU/g 0.6 1 8.79 2 8.79

3 8.79

4 8.85

5 9.19

6 9.06

Average 8.91

Std. Deviation 0.17

Appendix B – Survivor Data for Rapid Desiccation and Hydration Study

This appendix includes the raw Salmonella survivor data generated in the rapid desiccation and hydration study (Chapter 3). N/A indicates data for that point was unavailable due to experimental errors.

Table 8. Raw Salmonella survival data I60/F60

Rep Sample Time (s) Log CFU/g Log N/N0 REP 1 1 0 7.99 0.00 2 20 6.11 -1.88

3 40 3.30 N/A

4 60 6.21 -1.78

5 80 5.46 -2.52

6 100 5.69 -2.30

7 120 6.22 -1.77

8 140 5.75 -2.24

9 160 5.86 N/A

10 180 5.02 -2.97

REP 2 1 0 7.11 0.00 2 20 6.72 -0.39

3 40 5.76 N/A

4 60 6.12 -0.99

5 80 6.01 -1.09

6 100 5.67 -1.44

7 120 5.12 -1.98

8 140 6.01 -1.09

9 160 4.48 -2.62

10 180 4.68 -2.42

REP 3 1 0 5.69 0.00 2 20 6.51 0.83

3 40 6.80 1.11

4 60 5.46 -0.23

5 80 4.94 -0.74

6 100 4.46 -1.23

7 120 3.79 -1.90

8 140 4.91 -0.78

9 160 3.98 N/A

10 180 4.20 -1.48

Table 9. Raw Salmonella survivor data I30/F30

Rep Sample Time (s) Log CFU/g Log N/N0 REP 1 1 0 7.68 0.00 2 120 6.66 -1.02

3 240 7.56 -0.12

4 360 5.80 -1.87

5 480 6.24 -1.43

6 600 5.26 -2.42

7 720 6.10 -1.58

8 840 5.66 -2.01

9 960 5.60 -2.08

10 1080 5.61 -2.07

REP 2 1 0 8.11 0.00 2 120 6.56 -1.56

3 240 6.98 -1.13

4 360 6.63 -1.48

5 480 5.91 -2.20

6 600 5.88 -2.23

7 720 6.30 -1.81

8 840 5.56 -2.55

9 960 N/A N/A

10 1080 5.13 -2.98

REP 3 1 0 7.72 0.00 2 120 7.24 -0.48

3 240 7.15 -0.58

4 360 6.74 -0.98

5 480 6.13 -1.60

6 600 5.62 -2.10

7 720 6.27 -1.46

8 840 5.76 -1.96

9 960 4.08 -3.65

10 1080 4.04 -3.68

Table 10. Raw Salmonella survivor data I60/F30

Rep Sample Time (s) Log CFU/g Log N/N0 REP 1 1 0 7.71 0.00 2 60 6.88 -0.83

3 120 5.45 -2.26

4 180 5.58 -2.13

5 240 6.02 -1.69

6 300 5.27 -2.44

7 360 5.90 N/A

8 420 5.23 -2.48

9 480 5.34 N/A

10 540 5.93 -1.78

REP 2 1 0 6.86 0.00 2 60 7.46 N/A

3 120 5.26 -1.61

4 180 7.23 0.37

5 240 5.57 N/A

6 300 5.90 -0.96

7 360 5.63 -1.23

8 420 5.95 -0.91

9 480 5.69 -1.17

10 540 4.59 N/A

REP 3 1 0 7.54 0.00 2 60 6.45 -1.09

3 120 7.30 -0.24

4 180 6.82 -0.72

5 240 6.19 -1.35

6 300 5.97 -1.57

7 360 6.03 -1.52

8 420 3.64 -3.90

9 480 5.06 -2.49

10 540 4.02 -3.52

Table 11. Raw Salmonella survivor data I30/F60

Rep Sample Time (s) Log CFU/g Log N/N0 REP 1 1 0 6.66 0.00 2 30 5.24 -1.42

3 60 5.12 -1.54

4 90 4.32 -2.34

5 120 3.43 -3.23

6 150 6.38 -0.28

7 180 2.88 -3.79

8 210 1.70 -4.96

9 240 2.54 -4.12

10 270 2.40 -4.26

REP 2 1 0 6.76 0.00 2 30 5.88 -0.89

3 60 5.17 -1.59

4 90 4.35 -2.41

5 120 4.67 -2.10

6 150 4.05 -2.71

7 180 3.71 -3.05

8 210 4.29 -2.47

9 240 2.00 -4.76

10 270 N/A

REP 3 1 0 6.04 0.00 2 30 5.60 -0.44

3 60 5.09 -0.95

4 90 4.94 -1.10

5 120 4.21 -1.83

6 150 3.39 -2.65

7 180 2.98 -3.06

8 210 3.33 -2.71

9 240 3.18 -2.87

10 270 N/A N/A

Appendix C – Results of ANOVA

This appendix contains the results of ANOVA performed on the rapid desiccation or hydration study (Appendix B).

Table 12. ANOVA comparison of I60/F60 & I30/F30

SUMMARY Groups Count Sum Average Variance I60/F60 3 -0.0378 -0.0126 1.75E-06 I30/F30 3 -0.00724 -0.00241 5.52E-07

ANOVA Source of Variation SS df MS F P-value F crit Between Groups 0.000156 1 0.000156 135.2178 0.000313 7.708647 Within Groups 4.6E-06 4 1.15E-06 Total 0.00016 5

Table 13. ANOVA comparison of I60/F60 & I60/F30

SUMMARY Groups Count Sum Average Variance I60/F60 3 -0.0378 -0.0126 1.75E-06 I60/F30 3 -0.0111 -0.0037 5.33E-06

ANOVA Source of Variation SS df MS F P-value F crit Between Groups 0.000119 1 0.000119 33.54651 0.004417 7.708647 Within Groups 1.42E-05 4 3.54E-06 Total 0.000133 5

Table 14. ANOVA comparison of I60/F60 & I30/F60

SUMMARY Groups Count Sum Average Variance I60/F60 3 -0.0378 -0.0126 1.75E-06 I30/F60 3 -0.04368 -0.01456 1.12E-06

ANOVA Source of Variation SS df MS F P-value F crit Between Groups 5.75E-06 1 5.75E-06 4.011324 0.115742 7.708647 Within Groups 5.74E-06 4 1.43E-06 Total 1.15E-05 5

Table 15. ANOVA comparison of I30/F30 & I60/F30

SUMMARY Groups Count Sum Average Variance I30/F30 3 -0.00724 -0.00241 5.52E-07 I60/F30 3 -0.0111 -0.0037 5.33E-06

ANOVA Source of Variation SS df MS F P-value F crit Between Groups 2.48E-06 1 2.48E-06 0.844432 0.410131 7.708647 Within Groups 1.18E-05 4 2.94E-06 Total 1.42E-05 5

Table 16. ANOVA comparison of I30/F30 & I30/F60

SUMMARY Groups Count Sum Average Variance I30/F30 3 -0.00724 -0.00241 5.52E-07 I30/F60 3 -0.04368 -0.01456 1.12E-06

ANOVA Source of Variation SS df MS F P-value F crit Between Groups 0.000221 1 0.000221 264.8682 8.34E-05 7.708647 Within Groups 3.34E-06 4 8.35E-07 Total 0.000225 5

Table 17. ANOVA comparison of I60/F30 & I30/F60

SUMMARY Groups Count Sum Average Variance C 3 -0.0111 -0.0037 5.33E-06 D 3 -0.04368 -0.01456 1.12E-06

ANOVA Source of Variation SS df MS F P-value F crit Between Groups 0.000177 1 0.000177 54.82662 0.001775 7.708647 Within Groups 1.29E-05 4 3.23E-06 Total 0.00019 5

Appendix D – Photographs of Experimental Apparatus

This appendix includes pictures of systems utilized in this study.

Figure 9. Rapid desiccation system

Figure 10. Saturation chamber in rapid hydration system

Figure 11. Test cell filled with wheat flour that was used in thermal inactivation studies.

Figure 12. Equilibration chamber of the conditioning system

Appendix E – Rapid Desiccation and Hydration System Calibration Data

This appendix contains the data that were generated when the desiccation and hydration systems were calibrated, before inoculated trials were completed.

Table 18. Desiccation system calibration data

Sample Initial Final Difference 1 0.654 0.314 0.340 2 0.648 0.313 0.335 3 0.647 0.321 0.326 4 0.644 0.297 0.347 5 0.646 0.323 0.323 6 0.651 0.331 0.320 mean 0.648 0.317 0.332 std dev 0.004 0.012 0.011

Table 19. Hydration system calibration data

Sample Initial Final Difference 1 0.302 0.623 0.321 2 0.306 0.610 0.304 3 0.307 0.591 0.284 4 0.306 0.612 0.306 5 0.306 0.572 0.266 6 0.306 0.602 0.296 mean 0.306 0.602 0.296 std dev 0.002 0.018 0.019

Appendix F – Additional One Hour Hold Data

This appendix contains the raw survival count data from the additional one-hour hold experiment completed in the rapid desiccation/hydration study.

Table 20. One hour hold data (I60/F30/1H)

Rep Sample Time (s) Log CFU/g Log N/N0 Rep 1 1 0 8.55 0.00 2 60 8.70 0.15

3 120 7.54 -1.01

4 180 7.81 -0.74

5 240 7.31 -1.24

6 300 7.47 -1.08

7 360 7.25 -1.30

8 420 7.60 -0.95

9 480 6.01 -2.54

10 540 5.19 -3.36

Rep 2 1 0 8.94 0.00 2 60 8.41 -0.53

3 120 7.34 -1.60

4 180 7.39 -1.55

5 240 7.46 -1.48

6 300 7.47 -1.47

7 360 7.46 -1.48

8 420 7.33 -1.61

9 480 7.23 -1.71

10 540 7.28 -1.66

Rep 3 1 0 8.69 0.00 2 60 8.59 -0.10

3 120 7.29 -1.40

4 180 7.17 -1.52

5 240 7.32 -1.37

6 300 7.51 -1.17

7 360 7.17 -1.51

8 420 7.38 -1.30

9 480 7.22 -1.46

10 540 7.22 -1.46

Appendix G – Survivor Data for Factorial Modeling Study

This appendix contains raw survival count data generated from a full factorial modeling study encompassing thermal inactivation of Salmonella Enteritidis PT30 in wheat flour at three water activities (0.30, 0.45, 0.60) and three temperatures (75°C, 80°C, 85°C). N/A indicates data for that point was unavailable due to experimental problems/failures.

Table 21. Salmonella survival in wheat flour (aw = 0.30).

Water Time Temperature Sample Rep 1 Rep 2 Rep 3 Average Activity (s) 0.3 75 0 0 0.00 0.00 0.00 0.00 1 300 -0.67 -0.34 -0.31 -0.44 2 600 -0.99 -0.91 -0.36 -0.75 3 900 -1.45 -1.37 -0.88 -1.23 4 1200 -1.66 -1.65 -0.84 -1.39 5 1500 -2.08 -2.03 -1.09 -1.74 6 1800 -2.16 -2.19 -1.02 -1.79 7 2100 -2.74 -2.67 N/A -2.70 8 2400 -2.77 -2.48 N/A -2.62 0.3 80 0 0 0.00 0.00 0.00 0.00 1 240 -0.56 -0.96 -0.01 -0.51 2 480 -0.91 -1.51 -0.65 -1.02 3 720 -1.29 -1.85 -0.70 -1.28 4 960 -1.54 -2.74 -0.95 -1.74 5 1200 -1.65 -2.74 -1.70 -2.03 6 1440 -2.35 -2.74 -1.61 -2.23 7 1680 -2.48 -3.25 -1.69 -2.47 8 1920 -2.74 -4.59 -2.15 -3.16 0.3 85 0 0 0.00 0.00 0.00 0.00 1 120 -0.23 -0.78 -0.70 -0.57 2 240 -0.61 -0.59 -1.35 -0.85 3 360 -0.98 -1.48 -1.58 -1.35 4 480 -1.05 -1.69 -2.03 -1.59 5 600 -1.58 -2.18 -2.32 -2.03 6 720 -1.75 -2.18 -2.78 -2.24 7 840 -2.69 -2.98 -2.63 -2.77 8 960 -2.40 -3.42 -3.63 -3.15

Table 22. Salmonella survival in wheat flour (aw = 0.45).

Water Time Temperature Sample Rep 1 Rep 2 Rep 3 Average Activity (s) 0.45 75 0 0 0.00 0.00 0.00 0.00 1 420 -0.47 -1.07 -1.16 -0.90 2 840 -1.47 -1.60 -1.49 -1.52 3 1260 -1.53 -3.56 -2.21 -2.43 4 1680 -2.08 -3.89 -2.76 -2.91 5 2100 -2.38 -4.08 -4.07 -3.51 6 2520 -2.83 -4.85 -5.21 -4.30 7 2940 -3.43 -4.94 -6.25 -4.87 8 3360 -4.03 -6.45 -5.61 -5.36 0.45 80 0 0 0.00 0.00 0.00 0.00 1 180 -1.38 -0.47 -1.78 -1.21 2 360 -1.79 -0.55 -1.53 -1.29 3 540 -1.92 -1.18 -1.51 -1.54 4 720 -2.29 -1.58 -2.20 -2.02 5 900 -2.86 -3.05 -1.88 -2.59 6 1080 -3.53 -3.45 -2.63 -3.20 7 1260 -4.93 -3.65 -2.91 -3.83 8 1440 -4.90 -4.18 -4.23 -4.44 0.45 85 0 0 0.00 0.00 0.00 0.00 1 90 -0.50 -0.63 -1.08 -0.74 2 180 -1.16 -1.49 -2.28 -1.64 3 270 -1.45 -2.03 -3.17 -2.22 4 360 -2.12 -2.60 -4.48 -3.06 5 450 -3.32 -4.02 -4.95 -4.10 6 540 -3.74 -4.97 N/A -4.36 7 630 -4.26 -4.92 N/A -4.59 8 720 -4.80 -5.46 N/A -5.13

Table 23. Salmonella survival in wheat flour (aw = 0.60).

Water Time Temperature Sample Rep 1 Rep 2 Rep 3 Average Activity (s) 0.6 75 0 0 0.00 0.00 0.00 0.00 1 150 -0.56 -1.01 -0.74 -0.77 2 300 -1.48 -1.41 -1.28 -1.39 3 450 -3.55 -5.46 -2.02 -3.68 4 600 N/A -5.69 -2.78 -4.23 5 750 N/A -4.83 N/A -4.83 6 900 N/A -4.31 -4.65 -4.48 7 1050 N/A -7.16 -5.42 -6.29 8 1200 N/A N/A N/A N/A 0.6 80 0 0 0.00 0.00 0.00 0.00 1 30 -0.52 -0.65 -1.23 -0.80 2 60 -0.81 -1.00 -1.33 -1.04 3 90 -0.93 -1.20 -0.77 -0.97 4 120 -1.07 -1.68 -2.22 -1.66 5 150 -1.19 -2.27 -2.46 -1.98 6 180 -2.58 -2.46 -2.89 -2.64 7 210 -1.76 -2.57 -2.34 -2.22 8 240 -2.96 -3.48 -3.53 -3.32 0.6 85 0 0 0.00 0.00 0.00 0.00 1 20 0.07 -0.59 -0.71 -0.41 2 40 -1.84 -0.65 -0.50 -1.00 3 60 -1.97 -1.29 -1.23 -1.50 4 80 -2.02 -1.60 -1.76 -1.79 5 100 -4.00 -2.27 -3.71 -3.33 6 120 -2.22 -2.07 -2.86 -2.38 7 140 -4.12 -4.87 -4.11 -4.37 8 160 -2.28 -4.14 -4.25 -3.56

Appendix H – MATLAB Code for Log-Linear Model

This appendix contains the MATLAB code used to fit the log-linear primary model to the raw data from the modeling study.

%% Import Data data = xlsread('modelingdata.xlsx'); T1_60 = data(1:19,1:2); T2_60 = data(:,3:4); T3_60 = data(:,5:6); T1_45 = data(:,7:8); T2_45 = data(:,9:10); T3_45 = data(1:24,11:12); T1_30 = data(1:25,13:14); T2_30 = data(:,15:16); T3_30 = data(:,17:18);

% Sequentially arrange data T1_60=sortrows(T1_60,-2);T2_60=sortrows(T2_60,-2);T3_60=sortrows(T3_60,-2); T1_45=sortrows(T1_45,-2);T2_45=sortrows(T2_45,-2);T3_45=sortrows(T3_45,-2); T1_30=sortrows(T1_30,-2);T2_30=sortrows(T2_30,-2);T3_30=sortrows(T3_30,-2);

%% Estimate Betas M{1} = T1_60; M{2}= T2_60; M{3} = T3_60; M{4} = T1_45; M{5}= T2_45; M{6} = T3_45; M{7} = T1_30; M{8}= T2_30; M{9} = T3_30; k = length(M); for q = 1:k D = 1; beta0(1) = D; beta=beta0; t=M{1,q}(:,1); y=M{1,q}(:,2);

% Estimate Parameters [beta,R,J,COVB,mse] = nlinfit(t,y,@lin,beta0); D = beta(1);

% Statistics CI = nlparci(beta,R,'jacobian',J); % 95% CI [Rcorr,sigma] = corrcov(COVB); se = sigma; % Parameter Standard Error relerr = sigma./beta'; % Parameter Relative Error CorrCoef = Rcorr; % Correlation matrix condX = cond(J); rmse = sqrt(mse); res(q,:)=[q D CI(1,1) CI(1,2) se relerr rmse]; end disp(' Model # D CI Std. Error Rel. Error RMSE'); disp(res);

%% D value estimates D = res(:,2); Dmat = [D(1) D(2) D(3); D(4) D(5) D(6); D(7) D(8) D(9)];

%% Model Selection Data for m = 1:k p=1; t=M{1,m}(:,1); y=M{1,m}(:,2); n=length(t); beta = D(m); ynew = -t./beta; e = y-ynew; sum = e'*e; R = y.'*y; DR = R - sum; s2 = sum/(n-p); F = DR/s2; Fcritical=finv(0.95,1,n-p); K=p+1; AICc=n*log(sum/n)+2*K+2*K*(K+1)/(n-K-1); B(m,:)=[m p n-p sum s2 DR F Fcritical AICc]; end disp(' Model # p deg. of fr. R Mean sq. Delta R F Fcritical AICc'); disp(B);

function y = linear(beta,t) %first-order reaction equation, explicit form %beta are the parameters, and t are the independent variables values y=t/-beta(1); end

Appendix I – MATLAB Code for Weibull Model

This appendix contains the MATLAB code used to fit the Weibull primary model to the raw data from the modeling study.

%% Estimate D value M{1} = T1_60; M{2}= T2_60; M{3} = T3_60; M{4} = T1_45; M{5}= T2_45; M{6} = T3_45; M{7} = T1_30; M{8}= T2_30; M{9} = T3_30; k = length(M); for q = 1:k

D = 0.1; n =0.1; beta0(1)=D; beta0(2)=n; beta=beta0; t=M{1,q}(:,1); y=M{1,q}(:,2);

% Estimate Parameters [beta,R,J,COVB,mse] = nlinfit(t,y,@wbl,beta0); D = beta(1); n = beta(2);

% Statistics CI = nlparci(beta,R,'jacobian',J); % 95% CI [Rcorr,sigma] = corrcov(COVB); se = sigma; % Parameter Standard Error relerr = sigma./beta'; % Parameter Relative Error CorrCoef = Rcorr; % Correlation matrix condX = cond(J); rmse = sqrt(mse); res(q,:)=[q beta(1) se(1) beta(2) se(2) rmse]; end disp(' Model # Delta Std. Error n Std. Error RMSE'); disp(res); beta1=res(:,2); beta2=res(:,4); beta = [beta1 beta2];

%% Model Selection Data for m = 1:k p=2; t=M{1,m}(:,1); y=M{1,m}(:,2); n=length(t); beta1 = beta(m,1); beta2 = beta(m,2); ynew = -1.*(t./beta1).^beta2; e = y-ynew; sum = e'*e; R = y.'*y; DR = R - sum; s2 = sum/(n-p); F = DR/s2; Fcritical=finv(0.95,1,n-p); K=p+1; AICc=n*log(sum/n)+2*K+2*K*(K+1)/(n-K-1); B(m,:)=[m p n-p sum s2 DR F Fcritical AICc]; end disp(' Model # p deg. of fr. R Mean sq. Delta R F Fcritical AICc'); disp(B);

function y = weibull(beta,t) %first-order reaction equation, explicit form %beta are the parameters, and t are the independent variables values y=-1.*(t./beta(1)).^beta(2); %y=(t/-beta(1)) end

Appendix J – MATLAB Code for Secondary Models

This appendix contains the MATLAB code used to fit the secondary models to the raw data from the modeling study.

%% Import data data = xlsread('modelingdata2.xlsx'); t = data(:,1); yobs = data(:,2); T = data(:,3); aw = data(:,4); %t = t*60; n=length(t); aw_60 = data(1:73,:); aw_45 = data(74:151,:); aw_30 = data(152:230,:); T_75 = [data(1:19,:); data(74:100,:); data(152:176,:)]; T_80 = [data(20:46,:); data(101:127,:); data(177:203,:)]; T_85 = [data(47:73,:); data(128:151,:); data(204:230,:)]; %T=T-80; X = [t T aw];

%% SURFACE RESPONSE % Initial parameter guesses Tr = 80; beta0(1)= 800; % constant term beta0(2)= 4; % Temperature beta0(3)= 0.001; % Temperature^2 beta0(4)= 1000; % Water Activity beta0(5)= 1000; % Water Activity^2 beta0(6)= 1; % Interaction Term beta = beta0; p=length(beta0);

% Estimate Parameters [beta,resids,J,COVB,mse] = nlinfit(X,yobs,@surfresp,beta0); BSR = beta RMSE_SR = sqrt(mse) CI = nlparci(beta,resids,'jacobian',J) % 95% CI [Rcorr,sigma] = corrcov(COVB); se = sigma % Parameter Standard Error relerr = sigma./beta' % Parameter Relative Error CorrCoef = Rcorr; % Correlation matrix condX = cond(J);

% Use Estimated Parameters For Model Error D = beta(1)+(beta(2).*T)+(beta(3).*(T.^2))+(beta(4).*aw)+(beta(5).*(aw.^2))+(beta (6).*T.*aw); y = -t./D;

% AIC p = 6; e = yobs-y; sum = e'*e; R = y.'*y; DR = R - sum; s2 = sum/(n-p);

F = DR/s2; Fcritical = finv(0.95,1,n-p); K = p+1; AICc = n*log(sum/n)+(2*K)+(2*K*(K+1))/(n-K-1)

%% MAFART MODEL clear beta0 % Initial parameter guesses Tr = 80; beta0(1)= 10; % Dref beta0(2)= 20; % zT beta0(3)= 5; % zAw beta = beta0; p=length(beta0);

% Estimate Parameters [beta,resids,J,COVB,mse] = nlinfit(X,yobs,@modmafart,beta0); BM = beta RMSE_M = sqrt(mse) CI = nlparci(beta,resids,'jacobian',J); % 95% CI [Rcorr,sigma] = corrcov(COVB); se = sigma; % Parameter Standard Error relerr = sigma./beta'; % Parameter Relative Error CorrCoef = Rcorr; % Correlation matrix condX = cond(J);

% Use Estimated Parameters For Model Error D1 = log10(beta(1))-((T-Tr)./beta(2))-((aw-1)./beta(3)); D = 10.^(D1); y = -t./D;

% AIC p = 3; e = yobs-y; sum = e'*e; R = y.'*y; DR = R - sum; s2 = sum/(n-p); F = DR/s2; Fcritical = finv(0.95,1,n-p); K = p+1; AICc = n*log(sum/n)+2*K+2*K*(K+1)/(n-K-1)

% %% COMBINED EFFECTS clear beta0 % Initial parameter guesses Tr = 80; beta0(1)= 450; % Dref beta0(2)= 200; % Raw beta0(3)= 5; % zT beta = beta0; p=length(beta0);

% Estimate Parameters [beta,resids,J,COVB,mse] = nlinfit(X,yobs,@combeffects,beta0); b = beta 74

RMSE_CE = sqrt(mse) CI = nlparci(beta,resids,'jacobian',J); % 95% CI [Rcorr,sigma] = corrcov(COVB); se = sigma; % Parameter Standard Error relerr = sigma./beta'; % Parameter Relative Error CorrCoef = Rcorr; % Correlation matrix condX = cond(J);

% Use Estimated Parameters For Model Error D = (beta(1)+beta(2).*(0.6-aw)).*(10.^((Tr-T)/beta(3))); y = -t./D;

% AIC p =3; e = yobs-y; sum = e'*e; R = y.'*y; DR = R - sum; s2 = sum/(n-p); F = DR/s2; Fcritical = finv(0.95,1,n-p); K = p+1; AICc = n*log(sum/n)+2*K+2*K*(K+1)/(n-K-1)

function y = surfresp(beta,X) % This function represents the secondary surface response model global Tr t = X(:,1); T = X(:,2); aw = X(:,3);

D = beta(1)+(beta(2).*T)+(beta(3).*(T.^2))+(beta(4).*aw)+(beta(5).*(aw.^2))+(beta (6).*T.*aw); y = -t./D;

function y = modmafart(beta,X) % This function represents the secondary modified Mafart model global Tr t = X(:,1); T = X(:,2); aw = X(:,3);

D1 = log10(beta(1))-((T-Tr)./beta(2))-((aw-1)./beta(3)); D = 10.^(D1); y = -t./D;

function y = combeffects(beta,X) % This function represents the secondary combined effects model global Tr t = X(:,1); T = X(:,2); aw = X(:,3);

D = (beta(1)+beta(2).*(0.6-aw)).*(10.^((Tr-T)/beta(3))); y = -t./D;

Appendix K – Surface Plot of the Combined-Effects Model Response

This appendix contains the surface plot of the response of the D values when the combined- effects model is integrated with the log-linear model.

D values [min] 10

0 85 0.2 0.4 0.3 80 0.5 T [°C] 75 0.7 0.6 Water Activity

Figure 13. Surface response of the D values (min) with respect to aw and temperature (3D).

REFERENCES

Anderson, D. G., & Lucore, L. A. (2012). Validating the Reduction of Salmonella and Other Pathogens in Heat Processed Low-Moisture Foods. Alexandria, VA: Alliance for Innovation & Operational Excellence.

Anderson, N. M., Gradl, D., Pickens, S., Mishra, N., Li, H., & Keller, S. (2013). Method to determine differences in thermal tolerance of Salmonella serotypes at low water activity. Abstract #P1-105. Poster presented at the International Association for Food Protection Conference in Charlotte, NC.

Archer, J., Jervis, E. T., Bird, J., & Gaze, J. E. (1998). Heat resistance of Salmonella weltevreden in low-moisture environments. Journal of Food Protection, 61(8), 969-973.

Baranyi, J., & Roberts, T. A. (1994). A dynamic approach to predicting bacterial growth in food. International Journal of Food Microbiology, 23(3-4), 277-294.

Bell, C., & Kyriakides, A. (2002). Salmonella: A practical approach to the organism and its control in foods (First Edition ed.). Oxford, England: Blackwell Science LTD.

Beuchat, L. R., & Scouten, A. J. (2002). Combined effects of water activity, temperature and chemical treatments on the survival of Salmonella and Escherichia coli O157:H7 on alfalfa seeds. Journal of Applied Microbiology, 92(3), 382-395.

Bigelow, W. D., & Esty, J. R. (1920). The thermal death point in relation to time of typical thermophilic organisms. Journal of Infectious Diseases, 27, 602-617.

Burnett, S. L., Gehm, E. R., Weissinger, W. R., & Beuchat, L. R. (2000). Survival of Salmonella in peanut butter and peanut butter spread. Journal of Applied Microbiology, 89(3), 472- 477.

CDC. (2004). Outbreak of Salmonella serotype Enteritidis infection associated with raw almonds - United states and Canada, 2003-2004. Retrieved November 12, 2012, from http://www.cdc.gov/mmwr/preview/mmwrhtml/mm5322a8.htm

CDC. (2011). CDC Estimates of Foodborne Illness in the United States. Retrieved October 29, 2012, from http://www.cdc.gov/foodborneburden/2011-foodborne-estimates.html

CDC. (2014). List of Selected Multistate Foodborne Outbreak Investigations. Retrieved Jan 23, 2014, from http://www.cdc.gov/foodsafety/outbreaks/multistate-outbreaks/outbreaks- list.html#salmonella

Cerf, O. (1977). Tailing of survival curves of bacterial spores. Journal of Applied Bacteriology, 42(1), 1-19.

Cerf, O., Davey, K. R., & Sadoudi, A. K. (1996). Thermal inactivation of bacteria - A new predictive model for the combined effect of three environmental factors: Temperature, pH and water activity. Food Research International, 29(3-4), 219-226.

Chung, H. J., Birla, S. L., & Tang, J. (2008). Performance evaluation of aluminum test cell designed for determining the heat resistance of bacterial spores in foods. Lwt-Food Science and Technology, 41(8), 1351-1359.

Dolan, K. D., Valdramidis, V. P., & Mishra, D. K. (2013). Parameter estimation for dynamic microbial inactivation: which model, which precision? Food Control, 29(2), 401-408.

Doyle, M. E., & Mazzotta, A. S. (2000). Review of studies on the thermal resistance of salmonellae. Journal of Food Protection, 63(6), 779-795.

Elgazzar, F. E., & Marth, E. H. (1992). Foodborne disease – Investigative procedures and economic assessment. Journal of Environmental Health, 55(2), 24-27.

Farakos, S. M. S., Frank, J. F., & Schaffner, D. W. (2013). Modeling the influence of temperature, water activity and water mobility on the persistence of Salmonella in low- moisture foods. International Journal of Food Microbiology, 166(2), 280-293.

Farakos, S. M. S., Hicks, J. W., & Frank, J. E. (2014). Temperature resistance of Salmonella in low-water activity whey protein powder as influenced by salt content. Journal of Food Protection, 77(4), 631-634.

Farakos, S. M. S., Schaffner, D. W., & F., F. J. (2014). Predicting survival of Salmonella in low- water activity foods: An analysis of literature data. Journal of Food Protection. In press.

FDA. (2010). Recall: Products Containing Hydrolyzed Vegetable Protein. Retrieved October 29, 2012, from http://www.accessdata.fda.gov/scripts/HVPCP/

FDA. (2012). Bad Bug Book. In K. Lampel (Eds.), Foodborne Pathogenic Microorganisms and Natural Toxins

FDA. (2014). Background on the FDA Food Safety Modernization Act (FSMA). Retrieved April 22, 2014, from http://www.fda.gov/Food/GuidanceRegulation/FSMA/ucm239907.htm

Gaillard, S., Leguerinel, I., & Mafart, P. (1998). Model for combined effects of temperature, pH and water activity on thermal inactivation of Bacillus cereus spores. Journal of Food Science, 63(5), 887-889.

Geeraerd, A. H., Herremans, C. H., & Van Impe, J. F. (2000). Structural model requirements to describe microbial inactivation during a mild heat treatment. International Journal of Food Microbiology, 59(3), 185-209.

Gieraltowski, L., Julian, E., Pringle, J., Macdonald, K., Quilliam, D., Marsden-Haug, N., et al. (2013). Nationwide outbreak of Salmonella Montevideo infections associated with contaminated imported black and red pepper: warehouse membership cards provide critical clues to identify the source. [Article]. Epidemiology and Infection, 141(6), 1244- 1252.

GMA. (2009a). Annex to control of Salmonella in low-moisture foods. Retrieved from http://www.gmaonline.org/issues-policy/product-safety/food-and-product- safety/manufacturing-processing-and-regulatory-support/preventing-foodborne- illness/preventing-salmonella/

GMA. (2009b). Control of Salmonella in Low-Moisture Foods (Second Edition ed.). Retrieved from http://www.gmaonline.org/issues-policy/product-safety/food-and-product- safety/manufacturing-processing-and-regulatory-support/preventing-foodborne- illness/preventing-salmonella/

Goepfert, J. M., Iskander, I. K., & Amundson, C. H. (1970). Relation of heat resistance of salmonellae to water activity of environment. Applied Microbiology, 19(3), 429-433.

Gruzdev, N., Pinto, R., & Sela, S. (2011). Effect of desiccation on tolerance of Salmonella enterica to multiple stresses. Applied and Environmental Microbiology, 77(5), 1667- 1673.

He, Y. S., Guo, D. J., Yang, J. Y., Tortorello, M. L., & Zhang, W. (2011). Survival and heat resistance of Salmonella enterica and Escherichia coli O157:H7 in peanut butter. Applied and Environmental Microbiology, 77(23), 8434-8438.

Hiramatsu, R., Matsumoto, M., Sakae, K., & Miyazaki, Y. (2005). Ability of Shiga toxin- producing Escherichia coli and Salmonella spp. to survive in a desiccation model system and in dry foods. Applied and Environmental Microbiology, 71(11), 6657-6663.

Isaacs, S., Aramini, J., Ciebin, B., Farrar, J. A., Ahmed, R., Middleton, D., et al. (2005). An international outbreak of salmonellosis associated with raw almonds contaminated with a rare phage type of Salmonella enteritidis. Journal of Food Protection, 68(1), 191-198.

Janning, B., Tveld, P. H. I., Notermans, S., & Kramer, J. (1994). Resistance of bacterial strains to dry conditions – Use of anhydrous silica-gel in a desiccation model. Journal of Applied Bacteriology, 77(3), 319-324.

Jeong, S., Marks, B. P., & Orta-Ramirez, A. (2009). Thermal Inactivation Kinetics for Salmonella Enteritidis PT30 on Almonds Subjected to Moist-Air Convection Heating. Journal of Food Protection, 72(8), 1602-1609.

Jeong, S., Marks, B. P., Ryser, E. T., & Harte, J. B. (2012). The effect of X-ray irradiation on Salmonella inactivation and sensory quality of almonds and walnuts as a function of water activity. International Journal of Food Microbiology, 153(3), 365-371.

Juven, B. J., Cox, N. A., Bailey, J. S., Thomson, J. E., Charles, O. W., & Shutze, J. V. (1984). Survival of Salmonella in dry food and feed. Journal of Food Protection, 47(6), 445-448.

Keady, S., Briggs, G., Farrar, J., Mohle-Boetani, J. C., Oconnell, J., Werner, S. B., et al. (2004). Outbreak of Salmonella serotype enteritidis infections associated with raw almonds - United States and Canada, 2003-2004. Morbidity and Mortality Weekly Report, 53(22), 484-487.

Keller, S. E., VanDoren, J. M., Grasso, E. M., & Halik, L. A. (2013). Growth and survival of Salmonella in ground black pepper (Piper nigrum). Food Microbiology, 34(1), 182-188.

Kimber, M. A., Kaur, H., Wang, L. X., Danyluk, M. D., & Harris, L. J. (2012). Survival of Salmonella, Escherichia coli O157:H7, and Listeria monocytogenes on inoculated almonds and pistachios stored at -19, 4, and 24°C. Journal of Food Protection, 75(8), 1394-1403.

Lehmacher, A., Bockemuhl, J., & Aleksic, S. (1995). Nationwide outbreak of human salmonellosis in Germany due to contaminated paprika and paprika-powdered potato chips. Epidemiology and Infection, 115(3), 501-511.

Li, H. P., Fu, X. W., Bima, Y. G., Koontz, J., Megalis, C., Yang, F., et al. (2014). Effect of the local microenvironment on survival and thermal inactivation of Salmonella in low- and intermediate-moisture multi-ingredient foods. Journal of Food Protection, 77(1), 67-74.

Ma, L., Zhang, G. D., Gerner-Smidt, P., Mantripragada, V., Ezeoke, I., & Doyle, M. P. (2009). Thermal inactivation of Salmonella in peanut butter. Journal of Food Protection, 72(8), 1596-1601.

Mattick, K. L., Jorgensen, F., Wang, P., Pound, J., Vandeven, M. H., Ward, L. R., et al. (2001). Effect of challenge temperature and solute type on heat tolerance of Salmonella serovars at low water activity. Applied and Environmental Microbiology, 67(9), 4128-4136.

McCallum, L., Paine, S., Sexton, K., Dufour, M., Dyet, K., Wilson, M., et al. (2013). An outbreak of Salmonella Typhimurium Phage Type 42 associated with the consumption of raw flour. Foodborne Pathogens and Disease, 10(2), 159-164.

McKellar, R. C., & Lu, X. (2004). Modeling Microbial Responses in Food. Boca Raton: CRC Press.

Medus, C., Meyer, S., Smith, K., Jawahir, S., Miller, B., Viger, K., et al. (2009). Multistate outbreak of Salmonella infections associated with peanut butter and peanut butter- containing products - United States, 2008-2009. Morbidity and Mortality Weekly Report, 58(4), 85-90.

Motulsky, H., & Christopoulos, A. (2004). Fitting models to biological data using linear and nonlinear regression: a practical guide to curve fitting. New York: Oxford University Press.

Neter, J., Kutner, M., Wasserman, W., & Nachtsheim, C. (1996). Applied linear statistical models (4th ed. ed.). Chicago: McGraw-Hill/Irwin.

Park, E. J., Oh, S. W., & Kang, D. H. (2008). Fate of Salmonella Tennessee in peanut butter at 4 and 22°C. Journal of Food Science, 73(2), M82-M86.

Peleg, M. (2006). Advanced Quantitative Microbiology for Foods and Biosystems. Boca Raton: CRC Press.

Podolak, R., Enache, E., Stone, W., Black, D. G., & Elliott, P. H. (2010). Sources and risk factors for contamination, survival, persistence, and heat resistance of Salmonella in low- moisture foods. Journal of Food Protection, 73(10), 1919-1936.

Scallan, E., Hoekstra, R. M., Angulo, F. J., Tauxe, R. V., Widdowson, M. A., Roy, S. L., et al. (2011). Foodborne illness acquired in the United States-Major pathogens. Emerging Infectious Diseases, 17(1), 7-15.

Scharff, R. L. (2012). Economic burden from health losses due to foodborne illness in the United States. Journal of Food Protection, 75(1), 123-131.

Shachar, D., & Yaron, S. (2006). Heat tolerance of Salmonella enterica serovars Agona, Enteritidis, and Typhimurium in peanut butter. Journal of Food Protection, 69(11), 2687- 2691.

Sheth, A. N., Hoekstra, M., Patel, N., Ewald, G., Lord, C., Clarke, C., et al. (2011). A national outbreak of Salmonella serotype Tennessee infections from contaminated peanut butter: A new food vehicle for salmonellosis in the United States. Clinical Infectious Diseases, 53(4), 356-362.

Spector, M. P., & Kenyon, W. J. (2012). Resistance and survival strategies of Salmonella enterica to environmental stresses. Food Research International, 45(2), 455-481.

Sperber, W. H., & North American Millers (2007). Role of microbiological guidelines in the production and commercial use of milled cereal grains: A practical approach for the 21st century. Journal of Food Protection, 70(4), 1041-1053.

Uesugi, A. R., Danyluk, M. D., & Harris, L. J. (2006). Survival of Salmonella enteritidis phage type 30 on inoculated almonds stored at -20, 4, 23, and 35°C. Journal of Food Protection, 69(8), 1851-1857.

Usera, M. A., Echeita, A., Aladuena, A., Blanco, M. C., Reymundo, R., Prieto, M. I., et al. (1996). Interregional foodborne salmonellosis outbreak due to powdered infant formula contaminated with lactose-fermenting Salmonella virchow. European Journal of Epidemiology, 12(4), 377-381.

Valdramidis, V. P., Geeraerd, A. H., Gaze, J. E., Kondjoyan, A., Boyd, A. R., Shaw, H. L., et al. (2006). Quantitative description of Listeria monocytogenes inactivation kinetics with temperature and water activity as the influencing factors; model prediction and methodological validation on dynamic data. Journal of Food Engineering, 76(1), 79-88.

Vancauwenberge, J. E., Bothast, R. J., & Kwolek, W. F. (1981). Thermal inactivation of 8 Salmonella serotypes on dry corn flour. Applied and Environmental Microbiology, 42(4), 688-691.

Villa-Rojas, R., Tang, J. M., Wang, S. J., Gao, M. X., Kang, D. H., Mah, J. H., et al. (2013). Thermal inactivation of Salmonella Enteritidis PT 30 in almond kernels as influenced by water activity. Journal of Food Protection, 76(1), 26-32.